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/mm.h> 45 #include <linux/dma-mapping.h> 46 #include <linux/kref.h> 47 #include <linux/list.h> 48 #include <linux/rwsem.h> 49 #include <linux/scatterlist.h> 50 #include <linux/workqueue.h> 51 #include <linux/socket.h> 52 #include <linux/irq_poll.h> 53 #include <uapi/linux/if_ether.h> 54 #include <net/ipv6.h> 55 #include <net/ip.h> 56 #include <linux/string.h> 57 #include <linux/slab.h> 58 #include <linux/netdevice.h> 59 60 #include <linux/if_link.h> 61 #include <linux/atomic.h> 62 #include <linux/mmu_notifier.h> 63 #include <linux/uaccess.h> 64 #include <linux/cgroup_rdma.h> 65 #include <uapi/rdma/ib_user_verbs.h> 66 #include <rdma/restrack.h> 67 #include <uapi/rdma/rdma_user_ioctl.h> 68 #include <uapi/rdma/ib_user_ioctl_verbs.h> 69 70 #define IB_FW_VERSION_NAME_MAX ETHTOOL_FWVERS_LEN 71 72 extern struct workqueue_struct *ib_wq; 73 extern struct workqueue_struct *ib_comp_wq; 74 75 union ib_gid { 76 u8 raw[16]; 77 struct { 78 __be64 subnet_prefix; 79 __be64 interface_id; 80 } global; 81 }; 82 83 extern union ib_gid zgid; 84 85 enum ib_gid_type { 86 /* If link layer is Ethernet, this is RoCE V1 */ 87 IB_GID_TYPE_IB = 0, 88 IB_GID_TYPE_ROCE = 0, 89 IB_GID_TYPE_ROCE_UDP_ENCAP = 1, 90 IB_GID_TYPE_SIZE 91 }; 92 93 #define ROCE_V2_UDP_DPORT 4791 94 struct ib_gid_attr { 95 struct net_device *ndev; 96 struct ib_device *device; 97 enum ib_gid_type gid_type; 98 u16 index; 99 u8 port_num; 100 }; 101 102 enum rdma_node_type { 103 /* IB values map to NodeInfo:NodeType. */ 104 RDMA_NODE_IB_CA = 1, 105 RDMA_NODE_IB_SWITCH, 106 RDMA_NODE_IB_ROUTER, 107 RDMA_NODE_RNIC, 108 RDMA_NODE_USNIC, 109 RDMA_NODE_USNIC_UDP, 110 }; 111 112 enum { 113 /* set the local administered indication */ 114 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2, 115 }; 116 117 enum rdma_transport_type { 118 RDMA_TRANSPORT_IB, 119 RDMA_TRANSPORT_IWARP, 120 RDMA_TRANSPORT_USNIC, 121 RDMA_TRANSPORT_USNIC_UDP 122 }; 123 124 enum rdma_protocol_type { 125 RDMA_PROTOCOL_IB, 126 RDMA_PROTOCOL_IBOE, 127 RDMA_PROTOCOL_IWARP, 128 RDMA_PROTOCOL_USNIC_UDP 129 }; 130 131 __attribute_const__ enum rdma_transport_type 132 rdma_node_get_transport(enum rdma_node_type node_type); 133 134 enum rdma_network_type { 135 RDMA_NETWORK_IB, 136 RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB, 137 RDMA_NETWORK_IPV4, 138 RDMA_NETWORK_IPV6 139 }; 140 141 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type) 142 { 143 if (network_type == RDMA_NETWORK_IPV4 || 144 network_type == RDMA_NETWORK_IPV6) 145 return IB_GID_TYPE_ROCE_UDP_ENCAP; 146 147 /* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */ 148 return IB_GID_TYPE_IB; 149 } 150 151 static inline enum rdma_network_type ib_gid_to_network_type(enum ib_gid_type gid_type, 152 union ib_gid *gid) 153 { 154 if (gid_type == IB_GID_TYPE_IB) 155 return RDMA_NETWORK_IB; 156 157 if (ipv6_addr_v4mapped((struct in6_addr *)gid)) 158 return RDMA_NETWORK_IPV4; 159 else 160 return RDMA_NETWORK_IPV6; 161 } 162 163 enum rdma_link_layer { 164 IB_LINK_LAYER_UNSPECIFIED, 165 IB_LINK_LAYER_INFINIBAND, 166 IB_LINK_LAYER_ETHERNET, 167 }; 168 169 enum ib_device_cap_flags { 170 IB_DEVICE_RESIZE_MAX_WR = (1 << 0), 171 IB_DEVICE_BAD_PKEY_CNTR = (1 << 1), 172 IB_DEVICE_BAD_QKEY_CNTR = (1 << 2), 173 IB_DEVICE_RAW_MULTI = (1 << 3), 174 IB_DEVICE_AUTO_PATH_MIG = (1 << 4), 175 IB_DEVICE_CHANGE_PHY_PORT = (1 << 5), 176 IB_DEVICE_UD_AV_PORT_ENFORCE = (1 << 6), 177 IB_DEVICE_CURR_QP_STATE_MOD = (1 << 7), 178 IB_DEVICE_SHUTDOWN_PORT = (1 << 8), 179 /* Not in use, former INIT_TYPE = (1 << 9),*/ 180 IB_DEVICE_PORT_ACTIVE_EVENT = (1 << 10), 181 IB_DEVICE_SYS_IMAGE_GUID = (1 << 11), 182 IB_DEVICE_RC_RNR_NAK_GEN = (1 << 12), 183 IB_DEVICE_SRQ_RESIZE = (1 << 13), 184 IB_DEVICE_N_NOTIFY_CQ = (1 << 14), 185 186 /* 187 * This device supports a per-device lkey or stag that can be 188 * used without performing a memory registration for the local 189 * memory. Note that ULPs should never check this flag, but 190 * instead of use the local_dma_lkey flag in the ib_pd structure, 191 * which will always contain a usable lkey. 192 */ 193 IB_DEVICE_LOCAL_DMA_LKEY = (1 << 15), 194 /* Reserved, old SEND_W_INV = (1 << 16),*/ 195 IB_DEVICE_MEM_WINDOW = (1 << 17), 196 /* 197 * Devices should set IB_DEVICE_UD_IP_SUM if they support 198 * insertion of UDP and TCP checksum on outgoing UD IPoIB 199 * messages and can verify the validity of checksum for 200 * incoming messages. Setting this flag implies that the 201 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode. 202 */ 203 IB_DEVICE_UD_IP_CSUM = (1 << 18), 204 IB_DEVICE_UD_TSO = (1 << 19), 205 IB_DEVICE_XRC = (1 << 20), 206 207 /* 208 * This device supports the IB "base memory management extension", 209 * which includes support for fast registrations (IB_WR_REG_MR, 210 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should 211 * also be set by any iWarp device which must support FRs to comply 212 * to the iWarp verbs spec. iWarp devices also support the 213 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the 214 * stag. 215 */ 216 IB_DEVICE_MEM_MGT_EXTENSIONS = (1 << 21), 217 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1 << 22), 218 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1 << 23), 219 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1 << 24), 220 IB_DEVICE_RC_IP_CSUM = (1 << 25), 221 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */ 222 IB_DEVICE_RAW_IP_CSUM = (1 << 26), 223 /* 224 * Devices should set IB_DEVICE_CROSS_CHANNEL if they 225 * support execution of WQEs that involve synchronization 226 * of I/O operations with single completion queue managed 227 * by hardware. 228 */ 229 IB_DEVICE_CROSS_CHANNEL = (1 << 27), 230 IB_DEVICE_MANAGED_FLOW_STEERING = (1 << 29), 231 IB_DEVICE_SIGNATURE_HANDOVER = (1 << 30), 232 IB_DEVICE_ON_DEMAND_PAGING = (1ULL << 31), 233 IB_DEVICE_SG_GAPS_REG = (1ULL << 32), 234 IB_DEVICE_VIRTUAL_FUNCTION = (1ULL << 33), 235 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */ 236 IB_DEVICE_RAW_SCATTER_FCS = (1ULL << 34), 237 IB_DEVICE_RDMA_NETDEV_OPA_VNIC = (1ULL << 35), 238 /* The device supports padding incoming writes to cacheline. */ 239 IB_DEVICE_PCI_WRITE_END_PADDING = (1ULL << 36), 240 }; 241 242 enum ib_signature_prot_cap { 243 IB_PROT_T10DIF_TYPE_1 = 1, 244 IB_PROT_T10DIF_TYPE_2 = 1 << 1, 245 IB_PROT_T10DIF_TYPE_3 = 1 << 2, 246 }; 247 248 enum ib_signature_guard_cap { 249 IB_GUARD_T10DIF_CRC = 1, 250 IB_GUARD_T10DIF_CSUM = 1 << 1, 251 }; 252 253 enum ib_atomic_cap { 254 IB_ATOMIC_NONE, 255 IB_ATOMIC_HCA, 256 IB_ATOMIC_GLOB 257 }; 258 259 enum ib_odp_general_cap_bits { 260 IB_ODP_SUPPORT = 1 << 0, 261 IB_ODP_SUPPORT_IMPLICIT = 1 << 1, 262 }; 263 264 enum ib_odp_transport_cap_bits { 265 IB_ODP_SUPPORT_SEND = 1 << 0, 266 IB_ODP_SUPPORT_RECV = 1 << 1, 267 IB_ODP_SUPPORT_WRITE = 1 << 2, 268 IB_ODP_SUPPORT_READ = 1 << 3, 269 IB_ODP_SUPPORT_ATOMIC = 1 << 4, 270 }; 271 272 struct ib_odp_caps { 273 uint64_t general_caps; 274 struct { 275 uint32_t rc_odp_caps; 276 uint32_t uc_odp_caps; 277 uint32_t ud_odp_caps; 278 } per_transport_caps; 279 }; 280 281 struct ib_rss_caps { 282 /* Corresponding bit will be set if qp type from 283 * 'enum ib_qp_type' is supported, e.g. 284 * supported_qpts |= 1 << IB_QPT_UD 285 */ 286 u32 supported_qpts; 287 u32 max_rwq_indirection_tables; 288 u32 max_rwq_indirection_table_size; 289 }; 290 291 enum ib_tm_cap_flags { 292 /* Support tag matching on RC transport */ 293 IB_TM_CAP_RC = 1 << 0, 294 }; 295 296 struct ib_tm_caps { 297 /* Max size of RNDV header */ 298 u32 max_rndv_hdr_size; 299 /* Max number of entries in tag matching list */ 300 u32 max_num_tags; 301 /* From enum ib_tm_cap_flags */ 302 u32 flags; 303 /* Max number of outstanding list operations */ 304 u32 max_ops; 305 /* Max number of SGE in tag matching entry */ 306 u32 max_sge; 307 }; 308 309 struct ib_cq_init_attr { 310 unsigned int cqe; 311 int comp_vector; 312 u32 flags; 313 }; 314 315 enum ib_cq_attr_mask { 316 IB_CQ_MODERATE = 1 << 0, 317 }; 318 319 struct ib_cq_caps { 320 u16 max_cq_moderation_count; 321 u16 max_cq_moderation_period; 322 }; 323 324 struct ib_dm_mr_attr { 325 u64 length; 326 u64 offset; 327 u32 access_flags; 328 }; 329 330 struct ib_dm_alloc_attr { 331 u64 length; 332 u32 alignment; 333 u32 flags; 334 }; 335 336 struct ib_device_attr { 337 u64 fw_ver; 338 __be64 sys_image_guid; 339 u64 max_mr_size; 340 u64 page_size_cap; 341 u32 vendor_id; 342 u32 vendor_part_id; 343 u32 hw_ver; 344 int max_qp; 345 int max_qp_wr; 346 u64 device_cap_flags; 347 int max_sge; 348 int max_sge_rd; 349 int max_cq; 350 int max_cqe; 351 int max_mr; 352 int max_pd; 353 int max_qp_rd_atom; 354 int max_ee_rd_atom; 355 int max_res_rd_atom; 356 int max_qp_init_rd_atom; 357 int max_ee_init_rd_atom; 358 enum ib_atomic_cap atomic_cap; 359 enum ib_atomic_cap masked_atomic_cap; 360 int max_ee; 361 int max_rdd; 362 int max_mw; 363 int max_raw_ipv6_qp; 364 int max_raw_ethy_qp; 365 int max_mcast_grp; 366 int max_mcast_qp_attach; 367 int max_total_mcast_qp_attach; 368 int max_ah; 369 int max_fmr; 370 int max_map_per_fmr; 371 int max_srq; 372 int max_srq_wr; 373 int max_srq_sge; 374 unsigned int max_fast_reg_page_list_len; 375 u16 max_pkeys; 376 u8 local_ca_ack_delay; 377 int sig_prot_cap; 378 int sig_guard_cap; 379 struct ib_odp_caps odp_caps; 380 uint64_t timestamp_mask; 381 uint64_t hca_core_clock; /* in KHZ */ 382 struct ib_rss_caps rss_caps; 383 u32 max_wq_type_rq; 384 u32 raw_packet_caps; /* Use ib_raw_packet_caps enum */ 385 struct ib_tm_caps tm_caps; 386 struct ib_cq_caps cq_caps; 387 u64 max_dm_size; 388 }; 389 390 enum ib_mtu { 391 IB_MTU_256 = 1, 392 IB_MTU_512 = 2, 393 IB_MTU_1024 = 3, 394 IB_MTU_2048 = 4, 395 IB_MTU_4096 = 5 396 }; 397 398 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu) 399 { 400 switch (mtu) { 401 case IB_MTU_256: return 256; 402 case IB_MTU_512: return 512; 403 case IB_MTU_1024: return 1024; 404 case IB_MTU_2048: return 2048; 405 case IB_MTU_4096: return 4096; 406 default: return -1; 407 } 408 } 409 410 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu) 411 { 412 if (mtu >= 4096) 413 return IB_MTU_4096; 414 else if (mtu >= 2048) 415 return IB_MTU_2048; 416 else if (mtu >= 1024) 417 return IB_MTU_1024; 418 else if (mtu >= 512) 419 return IB_MTU_512; 420 else 421 return IB_MTU_256; 422 } 423 424 enum ib_port_state { 425 IB_PORT_NOP = 0, 426 IB_PORT_DOWN = 1, 427 IB_PORT_INIT = 2, 428 IB_PORT_ARMED = 3, 429 IB_PORT_ACTIVE = 4, 430 IB_PORT_ACTIVE_DEFER = 5 431 }; 432 433 enum ib_port_cap_flags { 434 IB_PORT_SM = 1 << 1, 435 IB_PORT_NOTICE_SUP = 1 << 2, 436 IB_PORT_TRAP_SUP = 1 << 3, 437 IB_PORT_OPT_IPD_SUP = 1 << 4, 438 IB_PORT_AUTO_MIGR_SUP = 1 << 5, 439 IB_PORT_SL_MAP_SUP = 1 << 6, 440 IB_PORT_MKEY_NVRAM = 1 << 7, 441 IB_PORT_PKEY_NVRAM = 1 << 8, 442 IB_PORT_LED_INFO_SUP = 1 << 9, 443 IB_PORT_SM_DISABLED = 1 << 10, 444 IB_PORT_SYS_IMAGE_GUID_SUP = 1 << 11, 445 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP = 1 << 12, 446 IB_PORT_EXTENDED_SPEEDS_SUP = 1 << 14, 447 IB_PORT_CM_SUP = 1 << 16, 448 IB_PORT_SNMP_TUNNEL_SUP = 1 << 17, 449 IB_PORT_REINIT_SUP = 1 << 18, 450 IB_PORT_DEVICE_MGMT_SUP = 1 << 19, 451 IB_PORT_VENDOR_CLASS_SUP = 1 << 20, 452 IB_PORT_DR_NOTICE_SUP = 1 << 21, 453 IB_PORT_CAP_MASK_NOTICE_SUP = 1 << 22, 454 IB_PORT_BOOT_MGMT_SUP = 1 << 23, 455 IB_PORT_LINK_LATENCY_SUP = 1 << 24, 456 IB_PORT_CLIENT_REG_SUP = 1 << 25, 457 IB_PORT_IP_BASED_GIDS = 1 << 26, 458 }; 459 460 enum ib_port_width { 461 IB_WIDTH_1X = 1, 462 IB_WIDTH_4X = 2, 463 IB_WIDTH_8X = 4, 464 IB_WIDTH_12X = 8 465 }; 466 467 static inline int ib_width_enum_to_int(enum ib_port_width width) 468 { 469 switch (width) { 470 case IB_WIDTH_1X: return 1; 471 case IB_WIDTH_4X: return 4; 472 case IB_WIDTH_8X: return 8; 473 case IB_WIDTH_12X: return 12; 474 default: return -1; 475 } 476 } 477 478 enum ib_port_speed { 479 IB_SPEED_SDR = 1, 480 IB_SPEED_DDR = 2, 481 IB_SPEED_QDR = 4, 482 IB_SPEED_FDR10 = 8, 483 IB_SPEED_FDR = 16, 484 IB_SPEED_EDR = 32, 485 IB_SPEED_HDR = 64 486 }; 487 488 /** 489 * struct rdma_hw_stats 490 * @lock - Mutex to protect parallel write access to lifespan and values 491 * of counters, which are 64bits and not guaranteeed to be written 492 * atomicaly on 32bits systems. 493 * @timestamp - Used by the core code to track when the last update was 494 * @lifespan - Used by the core code to determine how old the counters 495 * should be before being updated again. Stored in jiffies, defaults 496 * to 10 milliseconds, drivers can override the default be specifying 497 * their own value during their allocation routine. 498 * @name - Array of pointers to static names used for the counters in 499 * directory. 500 * @num_counters - How many hardware counters there are. If name is 501 * shorter than this number, a kernel oops will result. Driver authors 502 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters) 503 * in their code to prevent this. 504 * @value - Array of u64 counters that are accessed by the sysfs code and 505 * filled in by the drivers get_stats routine 506 */ 507 struct rdma_hw_stats { 508 struct mutex lock; /* Protect lifespan and values[] */ 509 unsigned long timestamp; 510 unsigned long lifespan; 511 const char * const *names; 512 int num_counters; 513 u64 value[]; 514 }; 515 516 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10 517 /** 518 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct 519 * for drivers. 520 * @names - Array of static const char * 521 * @num_counters - How many elements in array 522 * @lifespan - How many milliseconds between updates 523 */ 524 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct( 525 const char * const *names, int num_counters, 526 unsigned long lifespan) 527 { 528 struct rdma_hw_stats *stats; 529 530 stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64), 531 GFP_KERNEL); 532 if (!stats) 533 return NULL; 534 stats->names = names; 535 stats->num_counters = num_counters; 536 stats->lifespan = msecs_to_jiffies(lifespan); 537 538 return stats; 539 } 540 541 542 /* Define bits for the various functionality this port needs to be supported by 543 * the core. 544 */ 545 /* Management 0x00000FFF */ 546 #define RDMA_CORE_CAP_IB_MAD 0x00000001 547 #define RDMA_CORE_CAP_IB_SMI 0x00000002 548 #define RDMA_CORE_CAP_IB_CM 0x00000004 549 #define RDMA_CORE_CAP_IW_CM 0x00000008 550 #define RDMA_CORE_CAP_IB_SA 0x00000010 551 #define RDMA_CORE_CAP_OPA_MAD 0x00000020 552 553 /* Address format 0x000FF000 */ 554 #define RDMA_CORE_CAP_AF_IB 0x00001000 555 #define RDMA_CORE_CAP_ETH_AH 0x00002000 556 #define RDMA_CORE_CAP_OPA_AH 0x00004000 557 558 /* Protocol 0xFFF00000 */ 559 #define RDMA_CORE_CAP_PROT_IB 0x00100000 560 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000 561 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000 562 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000 563 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000 564 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000 565 566 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \ 567 | RDMA_CORE_CAP_IB_MAD \ 568 | RDMA_CORE_CAP_IB_SMI \ 569 | RDMA_CORE_CAP_IB_CM \ 570 | RDMA_CORE_CAP_IB_SA \ 571 | RDMA_CORE_CAP_AF_IB) 572 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \ 573 | RDMA_CORE_CAP_IB_MAD \ 574 | RDMA_CORE_CAP_IB_CM \ 575 | RDMA_CORE_CAP_AF_IB \ 576 | RDMA_CORE_CAP_ETH_AH) 577 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \ 578 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \ 579 | RDMA_CORE_CAP_IB_MAD \ 580 | RDMA_CORE_CAP_IB_CM \ 581 | RDMA_CORE_CAP_AF_IB \ 582 | RDMA_CORE_CAP_ETH_AH) 583 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \ 584 | RDMA_CORE_CAP_IW_CM) 585 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \ 586 | RDMA_CORE_CAP_OPA_MAD) 587 588 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET) 589 590 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC) 591 592 struct ib_port_attr { 593 u64 subnet_prefix; 594 enum ib_port_state state; 595 enum ib_mtu max_mtu; 596 enum ib_mtu active_mtu; 597 int gid_tbl_len; 598 u32 port_cap_flags; 599 u32 max_msg_sz; 600 u32 bad_pkey_cntr; 601 u32 qkey_viol_cntr; 602 u16 pkey_tbl_len; 603 u32 sm_lid; 604 u32 lid; 605 u8 lmc; 606 u8 max_vl_num; 607 u8 sm_sl; 608 u8 subnet_timeout; 609 u8 init_type_reply; 610 u8 active_width; 611 u8 active_speed; 612 u8 phys_state; 613 bool grh_required; 614 }; 615 616 enum ib_device_modify_flags { 617 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0, 618 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1 619 }; 620 621 #define IB_DEVICE_NODE_DESC_MAX 64 622 623 struct ib_device_modify { 624 u64 sys_image_guid; 625 char node_desc[IB_DEVICE_NODE_DESC_MAX]; 626 }; 627 628 enum ib_port_modify_flags { 629 IB_PORT_SHUTDOWN = 1, 630 IB_PORT_INIT_TYPE = (1<<2), 631 IB_PORT_RESET_QKEY_CNTR = (1<<3), 632 IB_PORT_OPA_MASK_CHG = (1<<4) 633 }; 634 635 struct ib_port_modify { 636 u32 set_port_cap_mask; 637 u32 clr_port_cap_mask; 638 u8 init_type; 639 }; 640 641 enum ib_event_type { 642 IB_EVENT_CQ_ERR, 643 IB_EVENT_QP_FATAL, 644 IB_EVENT_QP_REQ_ERR, 645 IB_EVENT_QP_ACCESS_ERR, 646 IB_EVENT_COMM_EST, 647 IB_EVENT_SQ_DRAINED, 648 IB_EVENT_PATH_MIG, 649 IB_EVENT_PATH_MIG_ERR, 650 IB_EVENT_DEVICE_FATAL, 651 IB_EVENT_PORT_ACTIVE, 652 IB_EVENT_PORT_ERR, 653 IB_EVENT_LID_CHANGE, 654 IB_EVENT_PKEY_CHANGE, 655 IB_EVENT_SM_CHANGE, 656 IB_EVENT_SRQ_ERR, 657 IB_EVENT_SRQ_LIMIT_REACHED, 658 IB_EVENT_QP_LAST_WQE_REACHED, 659 IB_EVENT_CLIENT_REREGISTER, 660 IB_EVENT_GID_CHANGE, 661 IB_EVENT_WQ_FATAL, 662 }; 663 664 const char *__attribute_const__ ib_event_msg(enum ib_event_type event); 665 666 struct ib_event { 667 struct ib_device *device; 668 union { 669 struct ib_cq *cq; 670 struct ib_qp *qp; 671 struct ib_srq *srq; 672 struct ib_wq *wq; 673 u8 port_num; 674 } element; 675 enum ib_event_type event; 676 }; 677 678 struct ib_event_handler { 679 struct ib_device *device; 680 void (*handler)(struct ib_event_handler *, struct ib_event *); 681 struct list_head list; 682 }; 683 684 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \ 685 do { \ 686 (_ptr)->device = _device; \ 687 (_ptr)->handler = _handler; \ 688 INIT_LIST_HEAD(&(_ptr)->list); \ 689 } while (0) 690 691 struct ib_global_route { 692 union ib_gid dgid; 693 u32 flow_label; 694 u8 sgid_index; 695 u8 hop_limit; 696 u8 traffic_class; 697 }; 698 699 struct ib_grh { 700 __be32 version_tclass_flow; 701 __be16 paylen; 702 u8 next_hdr; 703 u8 hop_limit; 704 union ib_gid sgid; 705 union ib_gid dgid; 706 }; 707 708 union rdma_network_hdr { 709 struct ib_grh ibgrh; 710 struct { 711 /* The IB spec states that if it's IPv4, the header 712 * is located in the last 20 bytes of the header. 713 */ 714 u8 reserved[20]; 715 struct iphdr roce4grh; 716 }; 717 }; 718 719 #define IB_QPN_MASK 0xFFFFFF 720 721 enum { 722 IB_MULTICAST_QPN = 0xffffff 723 }; 724 725 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF) 726 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000) 727 728 enum ib_ah_flags { 729 IB_AH_GRH = 1 730 }; 731 732 enum ib_rate { 733 IB_RATE_PORT_CURRENT = 0, 734 IB_RATE_2_5_GBPS = 2, 735 IB_RATE_5_GBPS = 5, 736 IB_RATE_10_GBPS = 3, 737 IB_RATE_20_GBPS = 6, 738 IB_RATE_30_GBPS = 4, 739 IB_RATE_40_GBPS = 7, 740 IB_RATE_60_GBPS = 8, 741 IB_RATE_80_GBPS = 9, 742 IB_RATE_120_GBPS = 10, 743 IB_RATE_14_GBPS = 11, 744 IB_RATE_56_GBPS = 12, 745 IB_RATE_112_GBPS = 13, 746 IB_RATE_168_GBPS = 14, 747 IB_RATE_25_GBPS = 15, 748 IB_RATE_100_GBPS = 16, 749 IB_RATE_200_GBPS = 17, 750 IB_RATE_300_GBPS = 18 751 }; 752 753 /** 754 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the 755 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be 756 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec. 757 * @rate: rate to convert. 758 */ 759 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate); 760 761 /** 762 * ib_rate_to_mbps - Convert the IB rate enum to Mbps. 763 * For example, IB_RATE_2_5_GBPS will be converted to 2500. 764 * @rate: rate to convert. 765 */ 766 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate); 767 768 769 /** 770 * enum ib_mr_type - memory region type 771 * @IB_MR_TYPE_MEM_REG: memory region that is used for 772 * normal registration 773 * @IB_MR_TYPE_SIGNATURE: memory region that is used for 774 * signature operations (data-integrity 775 * capable regions) 776 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to 777 * register any arbitrary sg lists (without 778 * the normal mr constraints - see 779 * ib_map_mr_sg) 780 */ 781 enum ib_mr_type { 782 IB_MR_TYPE_MEM_REG, 783 IB_MR_TYPE_SIGNATURE, 784 IB_MR_TYPE_SG_GAPS, 785 }; 786 787 /** 788 * Signature types 789 * IB_SIG_TYPE_NONE: Unprotected. 790 * IB_SIG_TYPE_T10_DIF: Type T10-DIF 791 */ 792 enum ib_signature_type { 793 IB_SIG_TYPE_NONE, 794 IB_SIG_TYPE_T10_DIF, 795 }; 796 797 /** 798 * Signature T10-DIF block-guard types 799 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules. 800 * IB_T10DIF_CSUM: Corresponds to IP checksum rules. 801 */ 802 enum ib_t10_dif_bg_type { 803 IB_T10DIF_CRC, 804 IB_T10DIF_CSUM 805 }; 806 807 /** 808 * struct ib_t10_dif_domain - Parameters specific for T10-DIF 809 * domain. 810 * @bg_type: T10-DIF block guard type (CRC|CSUM) 811 * @pi_interval: protection information interval. 812 * @bg: seed of guard computation. 813 * @app_tag: application tag of guard block 814 * @ref_tag: initial guard block reference tag. 815 * @ref_remap: Indicate wethear the reftag increments each block 816 * @app_escape: Indicate to skip block check if apptag=0xffff 817 * @ref_escape: Indicate to skip block check if reftag=0xffffffff 818 * @apptag_check_mask: check bitmask of application tag. 819 */ 820 struct ib_t10_dif_domain { 821 enum ib_t10_dif_bg_type bg_type; 822 u16 pi_interval; 823 u16 bg; 824 u16 app_tag; 825 u32 ref_tag; 826 bool ref_remap; 827 bool app_escape; 828 bool ref_escape; 829 u16 apptag_check_mask; 830 }; 831 832 /** 833 * struct ib_sig_domain - Parameters for signature domain 834 * @sig_type: specific signauture type 835 * @sig: union of all signature domain attributes that may 836 * be used to set domain layout. 837 */ 838 struct ib_sig_domain { 839 enum ib_signature_type sig_type; 840 union { 841 struct ib_t10_dif_domain dif; 842 } sig; 843 }; 844 845 /** 846 * struct ib_sig_attrs - Parameters for signature handover operation 847 * @check_mask: bitmask for signature byte check (8 bytes) 848 * @mem: memory domain layout desciptor. 849 * @wire: wire domain layout desciptor. 850 */ 851 struct ib_sig_attrs { 852 u8 check_mask; 853 struct ib_sig_domain mem; 854 struct ib_sig_domain wire; 855 }; 856 857 enum ib_sig_err_type { 858 IB_SIG_BAD_GUARD, 859 IB_SIG_BAD_REFTAG, 860 IB_SIG_BAD_APPTAG, 861 }; 862 863 /** 864 * struct ib_sig_err - signature error descriptor 865 */ 866 struct ib_sig_err { 867 enum ib_sig_err_type err_type; 868 u32 expected; 869 u32 actual; 870 u64 sig_err_offset; 871 u32 key; 872 }; 873 874 enum ib_mr_status_check { 875 IB_MR_CHECK_SIG_STATUS = 1, 876 }; 877 878 /** 879 * struct ib_mr_status - Memory region status container 880 * 881 * @fail_status: Bitmask of MR checks status. For each 882 * failed check a corresponding status bit is set. 883 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS 884 * failure. 885 */ 886 struct ib_mr_status { 887 u32 fail_status; 888 struct ib_sig_err sig_err; 889 }; 890 891 /** 892 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate 893 * enum. 894 * @mult: multiple to convert. 895 */ 896 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult); 897 898 enum rdma_ah_attr_type { 899 RDMA_AH_ATTR_TYPE_UNDEFINED, 900 RDMA_AH_ATTR_TYPE_IB, 901 RDMA_AH_ATTR_TYPE_ROCE, 902 RDMA_AH_ATTR_TYPE_OPA, 903 }; 904 905 struct ib_ah_attr { 906 u16 dlid; 907 u8 src_path_bits; 908 }; 909 910 struct roce_ah_attr { 911 u8 dmac[ETH_ALEN]; 912 }; 913 914 struct opa_ah_attr { 915 u32 dlid; 916 u8 src_path_bits; 917 bool make_grd; 918 }; 919 920 struct rdma_ah_attr { 921 struct ib_global_route grh; 922 u8 sl; 923 u8 static_rate; 924 u8 port_num; 925 u8 ah_flags; 926 enum rdma_ah_attr_type type; 927 union { 928 struct ib_ah_attr ib; 929 struct roce_ah_attr roce; 930 struct opa_ah_attr opa; 931 }; 932 }; 933 934 enum ib_wc_status { 935 IB_WC_SUCCESS, 936 IB_WC_LOC_LEN_ERR, 937 IB_WC_LOC_QP_OP_ERR, 938 IB_WC_LOC_EEC_OP_ERR, 939 IB_WC_LOC_PROT_ERR, 940 IB_WC_WR_FLUSH_ERR, 941 IB_WC_MW_BIND_ERR, 942 IB_WC_BAD_RESP_ERR, 943 IB_WC_LOC_ACCESS_ERR, 944 IB_WC_REM_INV_REQ_ERR, 945 IB_WC_REM_ACCESS_ERR, 946 IB_WC_REM_OP_ERR, 947 IB_WC_RETRY_EXC_ERR, 948 IB_WC_RNR_RETRY_EXC_ERR, 949 IB_WC_LOC_RDD_VIOL_ERR, 950 IB_WC_REM_INV_RD_REQ_ERR, 951 IB_WC_REM_ABORT_ERR, 952 IB_WC_INV_EECN_ERR, 953 IB_WC_INV_EEC_STATE_ERR, 954 IB_WC_FATAL_ERR, 955 IB_WC_RESP_TIMEOUT_ERR, 956 IB_WC_GENERAL_ERR 957 }; 958 959 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status); 960 961 enum ib_wc_opcode { 962 IB_WC_SEND, 963 IB_WC_RDMA_WRITE, 964 IB_WC_RDMA_READ, 965 IB_WC_COMP_SWAP, 966 IB_WC_FETCH_ADD, 967 IB_WC_LSO, 968 IB_WC_LOCAL_INV, 969 IB_WC_REG_MR, 970 IB_WC_MASKED_COMP_SWAP, 971 IB_WC_MASKED_FETCH_ADD, 972 /* 973 * Set value of IB_WC_RECV so consumers can test if a completion is a 974 * receive by testing (opcode & IB_WC_RECV). 975 */ 976 IB_WC_RECV = 1 << 7, 977 IB_WC_RECV_RDMA_WITH_IMM 978 }; 979 980 enum ib_wc_flags { 981 IB_WC_GRH = 1, 982 IB_WC_WITH_IMM = (1<<1), 983 IB_WC_WITH_INVALIDATE = (1<<2), 984 IB_WC_IP_CSUM_OK = (1<<3), 985 IB_WC_WITH_SMAC = (1<<4), 986 IB_WC_WITH_VLAN = (1<<5), 987 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6), 988 }; 989 990 struct ib_wc { 991 union { 992 u64 wr_id; 993 struct ib_cqe *wr_cqe; 994 }; 995 enum ib_wc_status status; 996 enum ib_wc_opcode opcode; 997 u32 vendor_err; 998 u32 byte_len; 999 struct ib_qp *qp; 1000 union { 1001 __be32 imm_data; 1002 u32 invalidate_rkey; 1003 } ex; 1004 u32 src_qp; 1005 u32 slid; 1006 int wc_flags; 1007 u16 pkey_index; 1008 u8 sl; 1009 u8 dlid_path_bits; 1010 u8 port_num; /* valid only for DR SMPs on switches */ 1011 u8 smac[ETH_ALEN]; 1012 u16 vlan_id; 1013 u8 network_hdr_type; 1014 }; 1015 1016 enum ib_cq_notify_flags { 1017 IB_CQ_SOLICITED = 1 << 0, 1018 IB_CQ_NEXT_COMP = 1 << 1, 1019 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP, 1020 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2, 1021 }; 1022 1023 enum ib_srq_type { 1024 IB_SRQT_BASIC, 1025 IB_SRQT_XRC, 1026 IB_SRQT_TM, 1027 }; 1028 1029 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type) 1030 { 1031 return srq_type == IB_SRQT_XRC || 1032 srq_type == IB_SRQT_TM; 1033 } 1034 1035 enum ib_srq_attr_mask { 1036 IB_SRQ_MAX_WR = 1 << 0, 1037 IB_SRQ_LIMIT = 1 << 1, 1038 }; 1039 1040 struct ib_srq_attr { 1041 u32 max_wr; 1042 u32 max_sge; 1043 u32 srq_limit; 1044 }; 1045 1046 struct ib_srq_init_attr { 1047 void (*event_handler)(struct ib_event *, void *); 1048 void *srq_context; 1049 struct ib_srq_attr attr; 1050 enum ib_srq_type srq_type; 1051 1052 struct { 1053 struct ib_cq *cq; 1054 union { 1055 struct { 1056 struct ib_xrcd *xrcd; 1057 } xrc; 1058 1059 struct { 1060 u32 max_num_tags; 1061 } tag_matching; 1062 }; 1063 } ext; 1064 }; 1065 1066 struct ib_qp_cap { 1067 u32 max_send_wr; 1068 u32 max_recv_wr; 1069 u32 max_send_sge; 1070 u32 max_recv_sge; 1071 u32 max_inline_data; 1072 1073 /* 1074 * Maximum number of rdma_rw_ctx structures in flight at a time. 1075 * ib_create_qp() will calculate the right amount of neededed WRs 1076 * and MRs based on this. 1077 */ 1078 u32 max_rdma_ctxs; 1079 }; 1080 1081 enum ib_sig_type { 1082 IB_SIGNAL_ALL_WR, 1083 IB_SIGNAL_REQ_WR 1084 }; 1085 1086 enum ib_qp_type { 1087 /* 1088 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries 1089 * here (and in that order) since the MAD layer uses them as 1090 * indices into a 2-entry table. 1091 */ 1092 IB_QPT_SMI, 1093 IB_QPT_GSI, 1094 1095 IB_QPT_RC, 1096 IB_QPT_UC, 1097 IB_QPT_UD, 1098 IB_QPT_RAW_IPV6, 1099 IB_QPT_RAW_ETHERTYPE, 1100 IB_QPT_RAW_PACKET = 8, 1101 IB_QPT_XRC_INI = 9, 1102 IB_QPT_XRC_TGT, 1103 IB_QPT_MAX, 1104 IB_QPT_DRIVER = 0xFF, 1105 /* Reserve a range for qp types internal to the low level driver. 1106 * These qp types will not be visible at the IB core layer, so the 1107 * IB_QPT_MAX usages should not be affected in the core layer 1108 */ 1109 IB_QPT_RESERVED1 = 0x1000, 1110 IB_QPT_RESERVED2, 1111 IB_QPT_RESERVED3, 1112 IB_QPT_RESERVED4, 1113 IB_QPT_RESERVED5, 1114 IB_QPT_RESERVED6, 1115 IB_QPT_RESERVED7, 1116 IB_QPT_RESERVED8, 1117 IB_QPT_RESERVED9, 1118 IB_QPT_RESERVED10, 1119 }; 1120 1121 enum ib_qp_create_flags { 1122 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0, 1123 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1, 1124 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2, 1125 IB_QP_CREATE_MANAGED_SEND = 1 << 3, 1126 IB_QP_CREATE_MANAGED_RECV = 1 << 4, 1127 IB_QP_CREATE_NETIF_QP = 1 << 5, 1128 IB_QP_CREATE_SIGNATURE_EN = 1 << 6, 1129 /* FREE = 1 << 7, */ 1130 IB_QP_CREATE_SCATTER_FCS = 1 << 8, 1131 IB_QP_CREATE_CVLAN_STRIPPING = 1 << 9, 1132 IB_QP_CREATE_SOURCE_QPN = 1 << 10, 1133 IB_QP_CREATE_PCI_WRITE_END_PADDING = 1 << 11, 1134 /* reserve bits 26-31 for low level drivers' internal use */ 1135 IB_QP_CREATE_RESERVED_START = 1 << 26, 1136 IB_QP_CREATE_RESERVED_END = 1 << 31, 1137 }; 1138 1139 /* 1140 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler 1141 * callback to destroy the passed in QP. 1142 */ 1143 1144 struct ib_qp_init_attr { 1145 void (*event_handler)(struct ib_event *, void *); 1146 void *qp_context; 1147 struct ib_cq *send_cq; 1148 struct ib_cq *recv_cq; 1149 struct ib_srq *srq; 1150 struct ib_xrcd *xrcd; /* XRC TGT QPs only */ 1151 struct ib_qp_cap cap; 1152 enum ib_sig_type sq_sig_type; 1153 enum ib_qp_type qp_type; 1154 enum ib_qp_create_flags create_flags; 1155 1156 /* 1157 * Only needed for special QP types, or when using the RW API. 1158 */ 1159 u8 port_num; 1160 struct ib_rwq_ind_table *rwq_ind_tbl; 1161 u32 source_qpn; 1162 }; 1163 1164 struct ib_qp_open_attr { 1165 void (*event_handler)(struct ib_event *, void *); 1166 void *qp_context; 1167 u32 qp_num; 1168 enum ib_qp_type qp_type; 1169 }; 1170 1171 enum ib_rnr_timeout { 1172 IB_RNR_TIMER_655_36 = 0, 1173 IB_RNR_TIMER_000_01 = 1, 1174 IB_RNR_TIMER_000_02 = 2, 1175 IB_RNR_TIMER_000_03 = 3, 1176 IB_RNR_TIMER_000_04 = 4, 1177 IB_RNR_TIMER_000_06 = 5, 1178 IB_RNR_TIMER_000_08 = 6, 1179 IB_RNR_TIMER_000_12 = 7, 1180 IB_RNR_TIMER_000_16 = 8, 1181 IB_RNR_TIMER_000_24 = 9, 1182 IB_RNR_TIMER_000_32 = 10, 1183 IB_RNR_TIMER_000_48 = 11, 1184 IB_RNR_TIMER_000_64 = 12, 1185 IB_RNR_TIMER_000_96 = 13, 1186 IB_RNR_TIMER_001_28 = 14, 1187 IB_RNR_TIMER_001_92 = 15, 1188 IB_RNR_TIMER_002_56 = 16, 1189 IB_RNR_TIMER_003_84 = 17, 1190 IB_RNR_TIMER_005_12 = 18, 1191 IB_RNR_TIMER_007_68 = 19, 1192 IB_RNR_TIMER_010_24 = 20, 1193 IB_RNR_TIMER_015_36 = 21, 1194 IB_RNR_TIMER_020_48 = 22, 1195 IB_RNR_TIMER_030_72 = 23, 1196 IB_RNR_TIMER_040_96 = 24, 1197 IB_RNR_TIMER_061_44 = 25, 1198 IB_RNR_TIMER_081_92 = 26, 1199 IB_RNR_TIMER_122_88 = 27, 1200 IB_RNR_TIMER_163_84 = 28, 1201 IB_RNR_TIMER_245_76 = 29, 1202 IB_RNR_TIMER_327_68 = 30, 1203 IB_RNR_TIMER_491_52 = 31 1204 }; 1205 1206 enum ib_qp_attr_mask { 1207 IB_QP_STATE = 1, 1208 IB_QP_CUR_STATE = (1<<1), 1209 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2), 1210 IB_QP_ACCESS_FLAGS = (1<<3), 1211 IB_QP_PKEY_INDEX = (1<<4), 1212 IB_QP_PORT = (1<<5), 1213 IB_QP_QKEY = (1<<6), 1214 IB_QP_AV = (1<<7), 1215 IB_QP_PATH_MTU = (1<<8), 1216 IB_QP_TIMEOUT = (1<<9), 1217 IB_QP_RETRY_CNT = (1<<10), 1218 IB_QP_RNR_RETRY = (1<<11), 1219 IB_QP_RQ_PSN = (1<<12), 1220 IB_QP_MAX_QP_RD_ATOMIC = (1<<13), 1221 IB_QP_ALT_PATH = (1<<14), 1222 IB_QP_MIN_RNR_TIMER = (1<<15), 1223 IB_QP_SQ_PSN = (1<<16), 1224 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17), 1225 IB_QP_PATH_MIG_STATE = (1<<18), 1226 IB_QP_CAP = (1<<19), 1227 IB_QP_DEST_QPN = (1<<20), 1228 IB_QP_RESERVED1 = (1<<21), 1229 IB_QP_RESERVED2 = (1<<22), 1230 IB_QP_RESERVED3 = (1<<23), 1231 IB_QP_RESERVED4 = (1<<24), 1232 IB_QP_RATE_LIMIT = (1<<25), 1233 }; 1234 1235 enum ib_qp_state { 1236 IB_QPS_RESET, 1237 IB_QPS_INIT, 1238 IB_QPS_RTR, 1239 IB_QPS_RTS, 1240 IB_QPS_SQD, 1241 IB_QPS_SQE, 1242 IB_QPS_ERR 1243 }; 1244 1245 enum ib_mig_state { 1246 IB_MIG_MIGRATED, 1247 IB_MIG_REARM, 1248 IB_MIG_ARMED 1249 }; 1250 1251 enum ib_mw_type { 1252 IB_MW_TYPE_1 = 1, 1253 IB_MW_TYPE_2 = 2 1254 }; 1255 1256 struct ib_qp_attr { 1257 enum ib_qp_state qp_state; 1258 enum ib_qp_state cur_qp_state; 1259 enum ib_mtu path_mtu; 1260 enum ib_mig_state path_mig_state; 1261 u32 qkey; 1262 u32 rq_psn; 1263 u32 sq_psn; 1264 u32 dest_qp_num; 1265 int qp_access_flags; 1266 struct ib_qp_cap cap; 1267 struct rdma_ah_attr ah_attr; 1268 struct rdma_ah_attr alt_ah_attr; 1269 u16 pkey_index; 1270 u16 alt_pkey_index; 1271 u8 en_sqd_async_notify; 1272 u8 sq_draining; 1273 u8 max_rd_atomic; 1274 u8 max_dest_rd_atomic; 1275 u8 min_rnr_timer; 1276 u8 port_num; 1277 u8 timeout; 1278 u8 retry_cnt; 1279 u8 rnr_retry; 1280 u8 alt_port_num; 1281 u8 alt_timeout; 1282 u32 rate_limit; 1283 }; 1284 1285 enum ib_wr_opcode { 1286 IB_WR_RDMA_WRITE, 1287 IB_WR_RDMA_WRITE_WITH_IMM, 1288 IB_WR_SEND, 1289 IB_WR_SEND_WITH_IMM, 1290 IB_WR_RDMA_READ, 1291 IB_WR_ATOMIC_CMP_AND_SWP, 1292 IB_WR_ATOMIC_FETCH_AND_ADD, 1293 IB_WR_LSO, 1294 IB_WR_SEND_WITH_INV, 1295 IB_WR_RDMA_READ_WITH_INV, 1296 IB_WR_LOCAL_INV, 1297 IB_WR_REG_MR, 1298 IB_WR_MASKED_ATOMIC_CMP_AND_SWP, 1299 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD, 1300 IB_WR_REG_SIG_MR, 1301 /* reserve values for low level drivers' internal use. 1302 * These values will not be used at all in the ib core layer. 1303 */ 1304 IB_WR_RESERVED1 = 0xf0, 1305 IB_WR_RESERVED2, 1306 IB_WR_RESERVED3, 1307 IB_WR_RESERVED4, 1308 IB_WR_RESERVED5, 1309 IB_WR_RESERVED6, 1310 IB_WR_RESERVED7, 1311 IB_WR_RESERVED8, 1312 IB_WR_RESERVED9, 1313 IB_WR_RESERVED10, 1314 }; 1315 1316 enum ib_send_flags { 1317 IB_SEND_FENCE = 1, 1318 IB_SEND_SIGNALED = (1<<1), 1319 IB_SEND_SOLICITED = (1<<2), 1320 IB_SEND_INLINE = (1<<3), 1321 IB_SEND_IP_CSUM = (1<<4), 1322 1323 /* reserve bits 26-31 for low level drivers' internal use */ 1324 IB_SEND_RESERVED_START = (1 << 26), 1325 IB_SEND_RESERVED_END = (1 << 31), 1326 }; 1327 1328 struct ib_sge { 1329 u64 addr; 1330 u32 length; 1331 u32 lkey; 1332 }; 1333 1334 struct ib_cqe { 1335 void (*done)(struct ib_cq *cq, struct ib_wc *wc); 1336 }; 1337 1338 struct ib_send_wr { 1339 struct ib_send_wr *next; 1340 union { 1341 u64 wr_id; 1342 struct ib_cqe *wr_cqe; 1343 }; 1344 struct ib_sge *sg_list; 1345 int num_sge; 1346 enum ib_wr_opcode opcode; 1347 int send_flags; 1348 union { 1349 __be32 imm_data; 1350 u32 invalidate_rkey; 1351 } ex; 1352 }; 1353 1354 struct ib_rdma_wr { 1355 struct ib_send_wr wr; 1356 u64 remote_addr; 1357 u32 rkey; 1358 }; 1359 1360 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr) 1361 { 1362 return container_of(wr, struct ib_rdma_wr, wr); 1363 } 1364 1365 struct ib_atomic_wr { 1366 struct ib_send_wr wr; 1367 u64 remote_addr; 1368 u64 compare_add; 1369 u64 swap; 1370 u64 compare_add_mask; 1371 u64 swap_mask; 1372 u32 rkey; 1373 }; 1374 1375 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr) 1376 { 1377 return container_of(wr, struct ib_atomic_wr, wr); 1378 } 1379 1380 struct ib_ud_wr { 1381 struct ib_send_wr wr; 1382 struct ib_ah *ah; 1383 void *header; 1384 int hlen; 1385 int mss; 1386 u32 remote_qpn; 1387 u32 remote_qkey; 1388 u16 pkey_index; /* valid for GSI only */ 1389 u8 port_num; /* valid for DR SMPs on switch only */ 1390 }; 1391 1392 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr) 1393 { 1394 return container_of(wr, struct ib_ud_wr, wr); 1395 } 1396 1397 struct ib_reg_wr { 1398 struct ib_send_wr wr; 1399 struct ib_mr *mr; 1400 u32 key; 1401 int access; 1402 }; 1403 1404 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr) 1405 { 1406 return container_of(wr, struct ib_reg_wr, wr); 1407 } 1408 1409 struct ib_sig_handover_wr { 1410 struct ib_send_wr wr; 1411 struct ib_sig_attrs *sig_attrs; 1412 struct ib_mr *sig_mr; 1413 int access_flags; 1414 struct ib_sge *prot; 1415 }; 1416 1417 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr) 1418 { 1419 return container_of(wr, struct ib_sig_handover_wr, wr); 1420 } 1421 1422 struct ib_recv_wr { 1423 struct ib_recv_wr *next; 1424 union { 1425 u64 wr_id; 1426 struct ib_cqe *wr_cqe; 1427 }; 1428 struct ib_sge *sg_list; 1429 int num_sge; 1430 }; 1431 1432 enum ib_access_flags { 1433 IB_ACCESS_LOCAL_WRITE = 1, 1434 IB_ACCESS_REMOTE_WRITE = (1<<1), 1435 IB_ACCESS_REMOTE_READ = (1<<2), 1436 IB_ACCESS_REMOTE_ATOMIC = (1<<3), 1437 IB_ACCESS_MW_BIND = (1<<4), 1438 IB_ZERO_BASED = (1<<5), 1439 IB_ACCESS_ON_DEMAND = (1<<6), 1440 IB_ACCESS_HUGETLB = (1<<7), 1441 }; 1442 1443 /* 1444 * XXX: these are apparently used for ->rereg_user_mr, no idea why they 1445 * are hidden here instead of a uapi header! 1446 */ 1447 enum ib_mr_rereg_flags { 1448 IB_MR_REREG_TRANS = 1, 1449 IB_MR_REREG_PD = (1<<1), 1450 IB_MR_REREG_ACCESS = (1<<2), 1451 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1) 1452 }; 1453 1454 struct ib_fmr_attr { 1455 int max_pages; 1456 int max_maps; 1457 u8 page_shift; 1458 }; 1459 1460 struct ib_umem; 1461 1462 enum rdma_remove_reason { 1463 /* Userspace requested uobject deletion. Call could fail */ 1464 RDMA_REMOVE_DESTROY, 1465 /* Context deletion. This call should delete the actual object itself */ 1466 RDMA_REMOVE_CLOSE, 1467 /* Driver is being hot-unplugged. This call should delete the actual object itself */ 1468 RDMA_REMOVE_DRIVER_REMOVE, 1469 /* Context is being cleaned-up, but commit was just completed */ 1470 RDMA_REMOVE_DURING_CLEANUP, 1471 }; 1472 1473 struct ib_rdmacg_object { 1474 #ifdef CONFIG_CGROUP_RDMA 1475 struct rdma_cgroup *cg; /* owner rdma cgroup */ 1476 #endif 1477 }; 1478 1479 struct ib_ucontext { 1480 struct ib_device *device; 1481 struct ib_uverbs_file *ufile; 1482 int closing; 1483 1484 /* locking the uobjects_list */ 1485 struct mutex uobjects_lock; 1486 struct list_head uobjects; 1487 /* protects cleanup process from other actions */ 1488 struct rw_semaphore cleanup_rwsem; 1489 enum rdma_remove_reason cleanup_reason; 1490 1491 struct pid *tgid; 1492 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING 1493 struct rb_root_cached umem_tree; 1494 /* 1495 * Protects .umem_rbroot and tree, as well as odp_mrs_count and 1496 * mmu notifiers registration. 1497 */ 1498 struct rw_semaphore umem_rwsem; 1499 void (*invalidate_range)(struct ib_umem *umem, 1500 unsigned long start, unsigned long end); 1501 1502 struct mmu_notifier mn; 1503 atomic_t notifier_count; 1504 /* A list of umems that don't have private mmu notifier counters yet. */ 1505 struct list_head no_private_counters; 1506 int odp_mrs_count; 1507 #endif 1508 1509 struct ib_rdmacg_object cg_obj; 1510 }; 1511 1512 struct ib_uobject { 1513 u64 user_handle; /* handle given to us by userspace */ 1514 struct ib_ucontext *context; /* associated user context */ 1515 void *object; /* containing object */ 1516 struct list_head list; /* link to context's list */ 1517 struct ib_rdmacg_object cg_obj; /* rdmacg object */ 1518 int id; /* index into kernel idr */ 1519 struct kref ref; 1520 atomic_t usecnt; /* protects exclusive access */ 1521 struct rcu_head rcu; /* kfree_rcu() overhead */ 1522 1523 const struct uverbs_obj_type *type; 1524 }; 1525 1526 struct ib_uobject_file { 1527 struct ib_uobject uobj; 1528 /* ufile contains the lock between context release and file close */ 1529 struct ib_uverbs_file *ufile; 1530 }; 1531 1532 struct ib_udata { 1533 const void __user *inbuf; 1534 void __user *outbuf; 1535 size_t inlen; 1536 size_t outlen; 1537 }; 1538 1539 struct ib_pd { 1540 u32 local_dma_lkey; 1541 u32 flags; 1542 struct ib_device *device; 1543 struct ib_uobject *uobject; 1544 atomic_t usecnt; /* count all resources */ 1545 1546 u32 unsafe_global_rkey; 1547 1548 /* 1549 * Implementation details of the RDMA core, don't use in drivers: 1550 */ 1551 struct ib_mr *__internal_mr; 1552 struct rdma_restrack_entry res; 1553 }; 1554 1555 struct ib_xrcd { 1556 struct ib_device *device; 1557 atomic_t usecnt; /* count all exposed resources */ 1558 struct inode *inode; 1559 1560 struct mutex tgt_qp_mutex; 1561 struct list_head tgt_qp_list; 1562 }; 1563 1564 struct ib_ah { 1565 struct ib_device *device; 1566 struct ib_pd *pd; 1567 struct ib_uobject *uobject; 1568 enum rdma_ah_attr_type type; 1569 }; 1570 1571 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context); 1572 1573 enum ib_poll_context { 1574 IB_POLL_DIRECT, /* caller context, no hw completions */ 1575 IB_POLL_SOFTIRQ, /* poll from softirq context */ 1576 IB_POLL_WORKQUEUE, /* poll from workqueue */ 1577 }; 1578 1579 struct ib_cq { 1580 struct ib_device *device; 1581 struct ib_uobject *uobject; 1582 ib_comp_handler comp_handler; 1583 void (*event_handler)(struct ib_event *, void *); 1584 void *cq_context; 1585 int cqe; 1586 atomic_t usecnt; /* count number of work queues */ 1587 enum ib_poll_context poll_ctx; 1588 struct ib_wc *wc; 1589 union { 1590 struct irq_poll iop; 1591 struct work_struct work; 1592 }; 1593 /* 1594 * Implementation details of the RDMA core, don't use in drivers: 1595 */ 1596 struct rdma_restrack_entry res; 1597 }; 1598 1599 struct ib_srq { 1600 struct ib_device *device; 1601 struct ib_pd *pd; 1602 struct ib_uobject *uobject; 1603 void (*event_handler)(struct ib_event *, void *); 1604 void *srq_context; 1605 enum ib_srq_type srq_type; 1606 atomic_t usecnt; 1607 1608 struct { 1609 struct ib_cq *cq; 1610 union { 1611 struct { 1612 struct ib_xrcd *xrcd; 1613 u32 srq_num; 1614 } xrc; 1615 }; 1616 } ext; 1617 }; 1618 1619 enum ib_raw_packet_caps { 1620 /* Strip cvlan from incoming packet and report it in the matching work 1621 * completion is supported. 1622 */ 1623 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = (1 << 0), 1624 /* Scatter FCS field of an incoming packet to host memory is supported. 1625 */ 1626 IB_RAW_PACKET_CAP_SCATTER_FCS = (1 << 1), 1627 /* Checksum offloads are supported (for both send and receive). */ 1628 IB_RAW_PACKET_CAP_IP_CSUM = (1 << 2), 1629 /* When a packet is received for an RQ with no receive WQEs, the 1630 * packet processing is delayed. 1631 */ 1632 IB_RAW_PACKET_CAP_DELAY_DROP = (1 << 3), 1633 }; 1634 1635 enum ib_wq_type { 1636 IB_WQT_RQ 1637 }; 1638 1639 enum ib_wq_state { 1640 IB_WQS_RESET, 1641 IB_WQS_RDY, 1642 IB_WQS_ERR 1643 }; 1644 1645 struct ib_wq { 1646 struct ib_device *device; 1647 struct ib_uobject *uobject; 1648 void *wq_context; 1649 void (*event_handler)(struct ib_event *, void *); 1650 struct ib_pd *pd; 1651 struct ib_cq *cq; 1652 u32 wq_num; 1653 enum ib_wq_state state; 1654 enum ib_wq_type wq_type; 1655 atomic_t usecnt; 1656 }; 1657 1658 enum ib_wq_flags { 1659 IB_WQ_FLAGS_CVLAN_STRIPPING = 1 << 0, 1660 IB_WQ_FLAGS_SCATTER_FCS = 1 << 1, 1661 IB_WQ_FLAGS_DELAY_DROP = 1 << 2, 1662 IB_WQ_FLAGS_PCI_WRITE_END_PADDING = 1 << 3, 1663 }; 1664 1665 struct ib_wq_init_attr { 1666 void *wq_context; 1667 enum ib_wq_type wq_type; 1668 u32 max_wr; 1669 u32 max_sge; 1670 struct ib_cq *cq; 1671 void (*event_handler)(struct ib_event *, void *); 1672 u32 create_flags; /* Use enum ib_wq_flags */ 1673 }; 1674 1675 enum ib_wq_attr_mask { 1676 IB_WQ_STATE = 1 << 0, 1677 IB_WQ_CUR_STATE = 1 << 1, 1678 IB_WQ_FLAGS = 1 << 2, 1679 }; 1680 1681 struct ib_wq_attr { 1682 enum ib_wq_state wq_state; 1683 enum ib_wq_state curr_wq_state; 1684 u32 flags; /* Use enum ib_wq_flags */ 1685 u32 flags_mask; /* Use enum ib_wq_flags */ 1686 }; 1687 1688 struct ib_rwq_ind_table { 1689 struct ib_device *device; 1690 struct ib_uobject *uobject; 1691 atomic_t usecnt; 1692 u32 ind_tbl_num; 1693 u32 log_ind_tbl_size; 1694 struct ib_wq **ind_tbl; 1695 }; 1696 1697 struct ib_rwq_ind_table_init_attr { 1698 u32 log_ind_tbl_size; 1699 /* Each entry is a pointer to Receive Work Queue */ 1700 struct ib_wq **ind_tbl; 1701 }; 1702 1703 enum port_pkey_state { 1704 IB_PORT_PKEY_NOT_VALID = 0, 1705 IB_PORT_PKEY_VALID = 1, 1706 IB_PORT_PKEY_LISTED = 2, 1707 }; 1708 1709 struct ib_qp_security; 1710 1711 struct ib_port_pkey { 1712 enum port_pkey_state state; 1713 u16 pkey_index; 1714 u8 port_num; 1715 struct list_head qp_list; 1716 struct list_head to_error_list; 1717 struct ib_qp_security *sec; 1718 }; 1719 1720 struct ib_ports_pkeys { 1721 struct ib_port_pkey main; 1722 struct ib_port_pkey alt; 1723 }; 1724 1725 struct ib_qp_security { 1726 struct ib_qp *qp; 1727 struct ib_device *dev; 1728 /* Hold this mutex when changing port and pkey settings. */ 1729 struct mutex mutex; 1730 struct ib_ports_pkeys *ports_pkeys; 1731 /* A list of all open shared QP handles. Required to enforce security 1732 * properly for all users of a shared QP. 1733 */ 1734 struct list_head shared_qp_list; 1735 void *security; 1736 bool destroying; 1737 atomic_t error_list_count; 1738 struct completion error_complete; 1739 int error_comps_pending; 1740 }; 1741 1742 /* 1743 * @max_write_sge: Maximum SGE elements per RDMA WRITE request. 1744 * @max_read_sge: Maximum SGE elements per RDMA READ request. 1745 */ 1746 struct ib_qp { 1747 struct ib_device *device; 1748 struct ib_pd *pd; 1749 struct ib_cq *send_cq; 1750 struct ib_cq *recv_cq; 1751 spinlock_t mr_lock; 1752 int mrs_used; 1753 struct list_head rdma_mrs; 1754 struct list_head sig_mrs; 1755 struct ib_srq *srq; 1756 struct ib_xrcd *xrcd; /* XRC TGT QPs only */ 1757 struct list_head xrcd_list; 1758 1759 /* count times opened, mcast attaches, flow attaches */ 1760 atomic_t usecnt; 1761 struct list_head open_list; 1762 struct ib_qp *real_qp; 1763 struct ib_uobject *uobject; 1764 void (*event_handler)(struct ib_event *, void *); 1765 void *qp_context; 1766 u32 qp_num; 1767 u32 max_write_sge; 1768 u32 max_read_sge; 1769 enum ib_qp_type qp_type; 1770 struct ib_rwq_ind_table *rwq_ind_tbl; 1771 struct ib_qp_security *qp_sec; 1772 u8 port; 1773 1774 /* 1775 * Implementation details of the RDMA core, don't use in drivers: 1776 */ 1777 struct rdma_restrack_entry res; 1778 }; 1779 1780 struct ib_dm { 1781 struct ib_device *device; 1782 u32 length; 1783 u32 flags; 1784 struct ib_uobject *uobject; 1785 atomic_t usecnt; 1786 }; 1787 1788 struct ib_mr { 1789 struct ib_device *device; 1790 struct ib_pd *pd; 1791 u32 lkey; 1792 u32 rkey; 1793 u64 iova; 1794 u64 length; 1795 unsigned int page_size; 1796 bool need_inval; 1797 union { 1798 struct ib_uobject *uobject; /* user */ 1799 struct list_head qp_entry; /* FR */ 1800 }; 1801 1802 struct ib_dm *dm; 1803 1804 /* 1805 * Implementation details of the RDMA core, don't use in drivers: 1806 */ 1807 struct rdma_restrack_entry res; 1808 }; 1809 1810 struct ib_mw { 1811 struct ib_device *device; 1812 struct ib_pd *pd; 1813 struct ib_uobject *uobject; 1814 u32 rkey; 1815 enum ib_mw_type type; 1816 }; 1817 1818 struct ib_fmr { 1819 struct ib_device *device; 1820 struct ib_pd *pd; 1821 struct list_head list; 1822 u32 lkey; 1823 u32 rkey; 1824 }; 1825 1826 /* Supported steering options */ 1827 enum ib_flow_attr_type { 1828 /* steering according to rule specifications */ 1829 IB_FLOW_ATTR_NORMAL = 0x0, 1830 /* default unicast and multicast rule - 1831 * receive all Eth traffic which isn't steered to any QP 1832 */ 1833 IB_FLOW_ATTR_ALL_DEFAULT = 0x1, 1834 /* default multicast rule - 1835 * receive all Eth multicast traffic which isn't steered to any QP 1836 */ 1837 IB_FLOW_ATTR_MC_DEFAULT = 0x2, 1838 /* sniffer rule - receive all port traffic */ 1839 IB_FLOW_ATTR_SNIFFER = 0x3 1840 }; 1841 1842 /* Supported steering header types */ 1843 enum ib_flow_spec_type { 1844 /* L2 headers*/ 1845 IB_FLOW_SPEC_ETH = 0x20, 1846 IB_FLOW_SPEC_IB = 0x22, 1847 /* L3 header*/ 1848 IB_FLOW_SPEC_IPV4 = 0x30, 1849 IB_FLOW_SPEC_IPV6 = 0x31, 1850 IB_FLOW_SPEC_ESP = 0x34, 1851 /* L4 headers*/ 1852 IB_FLOW_SPEC_TCP = 0x40, 1853 IB_FLOW_SPEC_UDP = 0x41, 1854 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50, 1855 IB_FLOW_SPEC_INNER = 0x100, 1856 /* Actions */ 1857 IB_FLOW_SPEC_ACTION_TAG = 0x1000, 1858 IB_FLOW_SPEC_ACTION_DROP = 0x1001, 1859 IB_FLOW_SPEC_ACTION_HANDLE = 0x1002, 1860 }; 1861 #define IB_FLOW_SPEC_LAYER_MASK 0xF0 1862 #define IB_FLOW_SPEC_SUPPORT_LAYERS 8 1863 1864 /* Flow steering rule priority is set according to it's domain. 1865 * Lower domain value means higher priority. 1866 */ 1867 enum ib_flow_domain { 1868 IB_FLOW_DOMAIN_USER, 1869 IB_FLOW_DOMAIN_ETHTOOL, 1870 IB_FLOW_DOMAIN_RFS, 1871 IB_FLOW_DOMAIN_NIC, 1872 IB_FLOW_DOMAIN_NUM /* Must be last */ 1873 }; 1874 1875 enum ib_flow_flags { 1876 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */ 1877 IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */ 1878 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 3 /* Must be last */ 1879 }; 1880 1881 struct ib_flow_eth_filter { 1882 u8 dst_mac[6]; 1883 u8 src_mac[6]; 1884 __be16 ether_type; 1885 __be16 vlan_tag; 1886 /* Must be last */ 1887 u8 real_sz[0]; 1888 }; 1889 1890 struct ib_flow_spec_eth { 1891 u32 type; 1892 u16 size; 1893 struct ib_flow_eth_filter val; 1894 struct ib_flow_eth_filter mask; 1895 }; 1896 1897 struct ib_flow_ib_filter { 1898 __be16 dlid; 1899 __u8 sl; 1900 /* Must be last */ 1901 u8 real_sz[0]; 1902 }; 1903 1904 struct ib_flow_spec_ib { 1905 u32 type; 1906 u16 size; 1907 struct ib_flow_ib_filter val; 1908 struct ib_flow_ib_filter mask; 1909 }; 1910 1911 /* IPv4 header flags */ 1912 enum ib_ipv4_flags { 1913 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */ 1914 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the 1915 last have this flag set */ 1916 }; 1917 1918 struct ib_flow_ipv4_filter { 1919 __be32 src_ip; 1920 __be32 dst_ip; 1921 u8 proto; 1922 u8 tos; 1923 u8 ttl; 1924 u8 flags; 1925 /* Must be last */ 1926 u8 real_sz[0]; 1927 }; 1928 1929 struct ib_flow_spec_ipv4 { 1930 u32 type; 1931 u16 size; 1932 struct ib_flow_ipv4_filter val; 1933 struct ib_flow_ipv4_filter mask; 1934 }; 1935 1936 struct ib_flow_ipv6_filter { 1937 u8 src_ip[16]; 1938 u8 dst_ip[16]; 1939 __be32 flow_label; 1940 u8 next_hdr; 1941 u8 traffic_class; 1942 u8 hop_limit; 1943 /* Must be last */ 1944 u8 real_sz[0]; 1945 }; 1946 1947 struct ib_flow_spec_ipv6 { 1948 u32 type; 1949 u16 size; 1950 struct ib_flow_ipv6_filter val; 1951 struct ib_flow_ipv6_filter mask; 1952 }; 1953 1954 struct ib_flow_tcp_udp_filter { 1955 __be16 dst_port; 1956 __be16 src_port; 1957 /* Must be last */ 1958 u8 real_sz[0]; 1959 }; 1960 1961 struct ib_flow_spec_tcp_udp { 1962 u32 type; 1963 u16 size; 1964 struct ib_flow_tcp_udp_filter val; 1965 struct ib_flow_tcp_udp_filter mask; 1966 }; 1967 1968 struct ib_flow_tunnel_filter { 1969 __be32 tunnel_id; 1970 u8 real_sz[0]; 1971 }; 1972 1973 /* ib_flow_spec_tunnel describes the Vxlan tunnel 1974 * the tunnel_id from val has the vni value 1975 */ 1976 struct ib_flow_spec_tunnel { 1977 u32 type; 1978 u16 size; 1979 struct ib_flow_tunnel_filter val; 1980 struct ib_flow_tunnel_filter mask; 1981 }; 1982 1983 struct ib_flow_esp_filter { 1984 __be32 spi; 1985 __be32 seq; 1986 /* Must be last */ 1987 u8 real_sz[0]; 1988 }; 1989 1990 struct ib_flow_spec_esp { 1991 u32 type; 1992 u16 size; 1993 struct ib_flow_esp_filter val; 1994 struct ib_flow_esp_filter mask; 1995 }; 1996 1997 struct ib_flow_spec_action_tag { 1998 enum ib_flow_spec_type type; 1999 u16 size; 2000 u32 tag_id; 2001 }; 2002 2003 struct ib_flow_spec_action_drop { 2004 enum ib_flow_spec_type type; 2005 u16 size; 2006 }; 2007 2008 struct ib_flow_spec_action_handle { 2009 enum ib_flow_spec_type type; 2010 u16 size; 2011 struct ib_flow_action *act; 2012 }; 2013 2014 union ib_flow_spec { 2015 struct { 2016 u32 type; 2017 u16 size; 2018 }; 2019 struct ib_flow_spec_eth eth; 2020 struct ib_flow_spec_ib ib; 2021 struct ib_flow_spec_ipv4 ipv4; 2022 struct ib_flow_spec_tcp_udp tcp_udp; 2023 struct ib_flow_spec_ipv6 ipv6; 2024 struct ib_flow_spec_tunnel tunnel; 2025 struct ib_flow_spec_esp esp; 2026 struct ib_flow_spec_action_tag flow_tag; 2027 struct ib_flow_spec_action_drop drop; 2028 struct ib_flow_spec_action_handle action; 2029 }; 2030 2031 struct ib_flow_attr { 2032 enum ib_flow_attr_type type; 2033 u16 size; 2034 u16 priority; 2035 u32 flags; 2036 u8 num_of_specs; 2037 u8 port; 2038 /* Following are the optional layers according to user request 2039 * struct ib_flow_spec_xxx 2040 * struct ib_flow_spec_yyy 2041 */ 2042 }; 2043 2044 struct ib_flow { 2045 struct ib_qp *qp; 2046 struct ib_uobject *uobject; 2047 }; 2048 2049 enum ib_flow_action_type { 2050 IB_FLOW_ACTION_UNSPECIFIED, 2051 IB_FLOW_ACTION_ESP = 1, 2052 }; 2053 2054 struct ib_flow_action_attrs_esp_keymats { 2055 enum ib_uverbs_flow_action_esp_keymat protocol; 2056 union { 2057 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm; 2058 } keymat; 2059 }; 2060 2061 struct ib_flow_action_attrs_esp_replays { 2062 enum ib_uverbs_flow_action_esp_replay protocol; 2063 union { 2064 struct ib_uverbs_flow_action_esp_replay_bmp bmp; 2065 } replay; 2066 }; 2067 2068 enum ib_flow_action_attrs_esp_flags { 2069 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags 2070 * This is done in order to share the same flags between user-space and 2071 * kernel and spare an unnecessary translation. 2072 */ 2073 2074 /* Kernel flags */ 2075 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32, 2076 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33, 2077 }; 2078 2079 struct ib_flow_spec_list { 2080 struct ib_flow_spec_list *next; 2081 union ib_flow_spec spec; 2082 }; 2083 2084 struct ib_flow_action_attrs_esp { 2085 struct ib_flow_action_attrs_esp_keymats *keymat; 2086 struct ib_flow_action_attrs_esp_replays *replay; 2087 struct ib_flow_spec_list *encap; 2088 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled. 2089 * Value of 0 is a valid value. 2090 */ 2091 u32 esn; 2092 u32 spi; 2093 u32 seq; 2094 u32 tfc_pad; 2095 /* Use enum ib_flow_action_attrs_esp_flags */ 2096 u64 flags; 2097 u64 hard_limit_pkts; 2098 }; 2099 2100 struct ib_flow_action { 2101 struct ib_device *device; 2102 struct ib_uobject *uobject; 2103 enum ib_flow_action_type type; 2104 atomic_t usecnt; 2105 }; 2106 2107 struct ib_mad_hdr; 2108 struct ib_grh; 2109 2110 enum ib_process_mad_flags { 2111 IB_MAD_IGNORE_MKEY = 1, 2112 IB_MAD_IGNORE_BKEY = 2, 2113 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY 2114 }; 2115 2116 enum ib_mad_result { 2117 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */ 2118 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */ 2119 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */ 2120 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */ 2121 }; 2122 2123 struct ib_port_cache { 2124 u64 subnet_prefix; 2125 struct ib_pkey_cache *pkey; 2126 struct ib_gid_table *gid; 2127 u8 lmc; 2128 enum ib_port_state port_state; 2129 }; 2130 2131 struct ib_cache { 2132 rwlock_t lock; 2133 struct ib_event_handler event_handler; 2134 struct ib_port_cache *ports; 2135 }; 2136 2137 struct iw_cm_verbs; 2138 2139 struct ib_port_immutable { 2140 int pkey_tbl_len; 2141 int gid_tbl_len; 2142 u32 core_cap_flags; 2143 u32 max_mad_size; 2144 }; 2145 2146 /* rdma netdev type - specifies protocol type */ 2147 enum rdma_netdev_t { 2148 RDMA_NETDEV_OPA_VNIC, 2149 RDMA_NETDEV_IPOIB, 2150 }; 2151 2152 /** 2153 * struct rdma_netdev - rdma netdev 2154 * For cases where netstack interfacing is required. 2155 */ 2156 struct rdma_netdev { 2157 void *clnt_priv; 2158 struct ib_device *hca; 2159 u8 port_num; 2160 2161 /* cleanup function must be specified */ 2162 void (*free_rdma_netdev)(struct net_device *netdev); 2163 2164 /* control functions */ 2165 void (*set_id)(struct net_device *netdev, int id); 2166 /* send packet */ 2167 int (*send)(struct net_device *dev, struct sk_buff *skb, 2168 struct ib_ah *address, u32 dqpn); 2169 /* multicast */ 2170 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca, 2171 union ib_gid *gid, u16 mlid, 2172 int set_qkey, u32 qkey); 2173 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca, 2174 union ib_gid *gid, u16 mlid); 2175 }; 2176 2177 struct ib_port_pkey_list { 2178 /* Lock to hold while modifying the list. */ 2179 spinlock_t list_lock; 2180 struct list_head pkey_list; 2181 }; 2182 2183 struct uverbs_attr_bundle; 2184 2185 struct ib_device { 2186 /* Do not access @dma_device directly from ULP nor from HW drivers. */ 2187 struct device *dma_device; 2188 2189 char name[IB_DEVICE_NAME_MAX]; 2190 2191 struct list_head event_handler_list; 2192 spinlock_t event_handler_lock; 2193 2194 spinlock_t client_data_lock; 2195 struct list_head core_list; 2196 /* Access to the client_data_list is protected by the client_data_lock 2197 * spinlock and the lists_rwsem read-write semaphore */ 2198 struct list_head client_data_list; 2199 2200 struct ib_cache cache; 2201 /** 2202 * port_immutable is indexed by port number 2203 */ 2204 struct ib_port_immutable *port_immutable; 2205 2206 int num_comp_vectors; 2207 2208 struct ib_port_pkey_list *port_pkey_list; 2209 2210 struct iw_cm_verbs *iwcm; 2211 2212 /** 2213 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the 2214 * driver initialized data. The struct is kfree()'ed by the sysfs 2215 * core when the device is removed. A lifespan of -1 in the return 2216 * struct tells the core to set a default lifespan. 2217 */ 2218 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device, 2219 u8 port_num); 2220 /** 2221 * get_hw_stats - Fill in the counter value(s) in the stats struct. 2222 * @index - The index in the value array we wish to have updated, or 2223 * num_counters if we want all stats updated 2224 * Return codes - 2225 * < 0 - Error, no counters updated 2226 * index - Updated the single counter pointed to by index 2227 * num_counters - Updated all counters (will reset the timestamp 2228 * and prevent further calls for lifespan milliseconds) 2229 * Drivers are allowed to update all counters in leiu of just the 2230 * one given in index at their option 2231 */ 2232 int (*get_hw_stats)(struct ib_device *device, 2233 struct rdma_hw_stats *stats, 2234 u8 port, int index); 2235 int (*query_device)(struct ib_device *device, 2236 struct ib_device_attr *device_attr, 2237 struct ib_udata *udata); 2238 int (*query_port)(struct ib_device *device, 2239 u8 port_num, 2240 struct ib_port_attr *port_attr); 2241 enum rdma_link_layer (*get_link_layer)(struct ib_device *device, 2242 u8 port_num); 2243 /* When calling get_netdev, the HW vendor's driver should return the 2244 * net device of device @device at port @port_num or NULL if such 2245 * a net device doesn't exist. The vendor driver should call dev_hold 2246 * on this net device. The HW vendor's device driver must guarantee 2247 * that this function returns NULL before the net device has finished 2248 * NETDEV_UNREGISTER state. 2249 */ 2250 struct net_device *(*get_netdev)(struct ib_device *device, 2251 u8 port_num); 2252 /* query_gid should be return GID value for @device, when @port_num 2253 * link layer is either IB or iWarp. It is no-op if @port_num port 2254 * is RoCE link layer. 2255 */ 2256 int (*query_gid)(struct ib_device *device, 2257 u8 port_num, int index, 2258 union ib_gid *gid); 2259 /* When calling add_gid, the HW vendor's driver should add the gid 2260 * of device of port at gid index available at @attr. Meta-info of 2261 * that gid (for example, the network device related to this gid) is 2262 * available at @attr. @context allows the HW vendor driver to store 2263 * extra information together with a GID entry. The HW vendor driver may 2264 * allocate memory to contain this information and store it in @context 2265 * when a new GID entry is written to. Params are consistent until the 2266 * next call of add_gid or delete_gid. The function should return 0 on 2267 * success or error otherwise. The function could be called 2268 * concurrently for different ports. This function is only called when 2269 * roce_gid_table is used. 2270 */ 2271 int (*add_gid)(const union ib_gid *gid, 2272 const struct ib_gid_attr *attr, 2273 void **context); 2274 /* When calling del_gid, the HW vendor's driver should delete the 2275 * gid of device @device at gid index gid_index of port port_num 2276 * available in @attr. 2277 * Upon the deletion of a GID entry, the HW vendor must free any 2278 * allocated memory. The caller will clear @context afterwards. 2279 * This function is only called when roce_gid_table is used. 2280 */ 2281 int (*del_gid)(const struct ib_gid_attr *attr, 2282 void **context); 2283 int (*query_pkey)(struct ib_device *device, 2284 u8 port_num, u16 index, u16 *pkey); 2285 int (*modify_device)(struct ib_device *device, 2286 int device_modify_mask, 2287 struct ib_device_modify *device_modify); 2288 int (*modify_port)(struct ib_device *device, 2289 u8 port_num, int port_modify_mask, 2290 struct ib_port_modify *port_modify); 2291 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device, 2292 struct ib_udata *udata); 2293 int (*dealloc_ucontext)(struct ib_ucontext *context); 2294 int (*mmap)(struct ib_ucontext *context, 2295 struct vm_area_struct *vma); 2296 struct ib_pd * (*alloc_pd)(struct ib_device *device, 2297 struct ib_ucontext *context, 2298 struct ib_udata *udata); 2299 int (*dealloc_pd)(struct ib_pd *pd); 2300 struct ib_ah * (*create_ah)(struct ib_pd *pd, 2301 struct rdma_ah_attr *ah_attr, 2302 struct ib_udata *udata); 2303 int (*modify_ah)(struct ib_ah *ah, 2304 struct rdma_ah_attr *ah_attr); 2305 int (*query_ah)(struct ib_ah *ah, 2306 struct rdma_ah_attr *ah_attr); 2307 int (*destroy_ah)(struct ib_ah *ah); 2308 struct ib_srq * (*create_srq)(struct ib_pd *pd, 2309 struct ib_srq_init_attr *srq_init_attr, 2310 struct ib_udata *udata); 2311 int (*modify_srq)(struct ib_srq *srq, 2312 struct ib_srq_attr *srq_attr, 2313 enum ib_srq_attr_mask srq_attr_mask, 2314 struct ib_udata *udata); 2315 int (*query_srq)(struct ib_srq *srq, 2316 struct ib_srq_attr *srq_attr); 2317 int (*destroy_srq)(struct ib_srq *srq); 2318 int (*post_srq_recv)(struct ib_srq *srq, 2319 struct ib_recv_wr *recv_wr, 2320 struct ib_recv_wr **bad_recv_wr); 2321 struct ib_qp * (*create_qp)(struct ib_pd *pd, 2322 struct ib_qp_init_attr *qp_init_attr, 2323 struct ib_udata *udata); 2324 int (*modify_qp)(struct ib_qp *qp, 2325 struct ib_qp_attr *qp_attr, 2326 int qp_attr_mask, 2327 struct ib_udata *udata); 2328 int (*query_qp)(struct ib_qp *qp, 2329 struct ib_qp_attr *qp_attr, 2330 int qp_attr_mask, 2331 struct ib_qp_init_attr *qp_init_attr); 2332 int (*destroy_qp)(struct ib_qp *qp); 2333 int (*post_send)(struct ib_qp *qp, 2334 struct ib_send_wr *send_wr, 2335 struct ib_send_wr **bad_send_wr); 2336 int (*post_recv)(struct ib_qp *qp, 2337 struct ib_recv_wr *recv_wr, 2338 struct ib_recv_wr **bad_recv_wr); 2339 struct ib_cq * (*create_cq)(struct ib_device *device, 2340 const struct ib_cq_init_attr *attr, 2341 struct ib_ucontext *context, 2342 struct ib_udata *udata); 2343 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, 2344 u16 cq_period); 2345 int (*destroy_cq)(struct ib_cq *cq); 2346 int (*resize_cq)(struct ib_cq *cq, int cqe, 2347 struct ib_udata *udata); 2348 int (*poll_cq)(struct ib_cq *cq, int num_entries, 2349 struct ib_wc *wc); 2350 int (*peek_cq)(struct ib_cq *cq, int wc_cnt); 2351 int (*req_notify_cq)(struct ib_cq *cq, 2352 enum ib_cq_notify_flags flags); 2353 int (*req_ncomp_notif)(struct ib_cq *cq, 2354 int wc_cnt); 2355 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd, 2356 int mr_access_flags); 2357 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd, 2358 u64 start, u64 length, 2359 u64 virt_addr, 2360 int mr_access_flags, 2361 struct ib_udata *udata); 2362 int (*rereg_user_mr)(struct ib_mr *mr, 2363 int flags, 2364 u64 start, u64 length, 2365 u64 virt_addr, 2366 int mr_access_flags, 2367 struct ib_pd *pd, 2368 struct ib_udata *udata); 2369 int (*dereg_mr)(struct ib_mr *mr); 2370 struct ib_mr * (*alloc_mr)(struct ib_pd *pd, 2371 enum ib_mr_type mr_type, 2372 u32 max_num_sg); 2373 int (*map_mr_sg)(struct ib_mr *mr, 2374 struct scatterlist *sg, 2375 int sg_nents, 2376 unsigned int *sg_offset); 2377 struct ib_mw * (*alloc_mw)(struct ib_pd *pd, 2378 enum ib_mw_type type, 2379 struct ib_udata *udata); 2380 int (*dealloc_mw)(struct ib_mw *mw); 2381 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd, 2382 int mr_access_flags, 2383 struct ib_fmr_attr *fmr_attr); 2384 int (*map_phys_fmr)(struct ib_fmr *fmr, 2385 u64 *page_list, int list_len, 2386 u64 iova); 2387 int (*unmap_fmr)(struct list_head *fmr_list); 2388 int (*dealloc_fmr)(struct ib_fmr *fmr); 2389 int (*attach_mcast)(struct ib_qp *qp, 2390 union ib_gid *gid, 2391 u16 lid); 2392 int (*detach_mcast)(struct ib_qp *qp, 2393 union ib_gid *gid, 2394 u16 lid); 2395 int (*process_mad)(struct ib_device *device, 2396 int process_mad_flags, 2397 u8 port_num, 2398 const struct ib_wc *in_wc, 2399 const struct ib_grh *in_grh, 2400 const struct ib_mad_hdr *in_mad, 2401 size_t in_mad_size, 2402 struct ib_mad_hdr *out_mad, 2403 size_t *out_mad_size, 2404 u16 *out_mad_pkey_index); 2405 struct ib_xrcd * (*alloc_xrcd)(struct ib_device *device, 2406 struct ib_ucontext *ucontext, 2407 struct ib_udata *udata); 2408 int (*dealloc_xrcd)(struct ib_xrcd *xrcd); 2409 struct ib_flow * (*create_flow)(struct ib_qp *qp, 2410 struct ib_flow_attr 2411 *flow_attr, 2412 int domain); 2413 int (*destroy_flow)(struct ib_flow *flow_id); 2414 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask, 2415 struct ib_mr_status *mr_status); 2416 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext); 2417 void (*drain_rq)(struct ib_qp *qp); 2418 void (*drain_sq)(struct ib_qp *qp); 2419 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port, 2420 int state); 2421 int (*get_vf_config)(struct ib_device *device, int vf, u8 port, 2422 struct ifla_vf_info *ivf); 2423 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port, 2424 struct ifla_vf_stats *stats); 2425 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid, 2426 int type); 2427 struct ib_wq * (*create_wq)(struct ib_pd *pd, 2428 struct ib_wq_init_attr *init_attr, 2429 struct ib_udata *udata); 2430 int (*destroy_wq)(struct ib_wq *wq); 2431 int (*modify_wq)(struct ib_wq *wq, 2432 struct ib_wq_attr *attr, 2433 u32 wq_attr_mask, 2434 struct ib_udata *udata); 2435 struct ib_rwq_ind_table * (*create_rwq_ind_table)(struct ib_device *device, 2436 struct ib_rwq_ind_table_init_attr *init_attr, 2437 struct ib_udata *udata); 2438 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table); 2439 struct ib_flow_action * (*create_flow_action_esp)(struct ib_device *device, 2440 const struct ib_flow_action_attrs_esp *attr, 2441 struct uverbs_attr_bundle *attrs); 2442 int (*destroy_flow_action)(struct ib_flow_action *action); 2443 int (*modify_flow_action_esp)(struct ib_flow_action *action, 2444 const struct ib_flow_action_attrs_esp *attr, 2445 struct uverbs_attr_bundle *attrs); 2446 struct ib_dm * (*alloc_dm)(struct ib_device *device, 2447 struct ib_ucontext *context, 2448 struct ib_dm_alloc_attr *attr, 2449 struct uverbs_attr_bundle *attrs); 2450 int (*dealloc_dm)(struct ib_dm *dm); 2451 struct ib_mr * (*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm, 2452 struct ib_dm_mr_attr *attr, 2453 struct uverbs_attr_bundle *attrs); 2454 /** 2455 * rdma netdev operation 2456 * 2457 * Driver implementing alloc_rdma_netdev must return -EOPNOTSUPP if it 2458 * doesn't support the specified rdma netdev type. 2459 */ 2460 struct net_device *(*alloc_rdma_netdev)( 2461 struct ib_device *device, 2462 u8 port_num, 2463 enum rdma_netdev_t type, 2464 const char *name, 2465 unsigned char name_assign_type, 2466 void (*setup)(struct net_device *)); 2467 2468 struct module *owner; 2469 struct device dev; 2470 struct kobject *ports_parent; 2471 struct list_head port_list; 2472 2473 enum { 2474 IB_DEV_UNINITIALIZED, 2475 IB_DEV_REGISTERED, 2476 IB_DEV_UNREGISTERED 2477 } reg_state; 2478 2479 int uverbs_abi_ver; 2480 u64 uverbs_cmd_mask; 2481 u64 uverbs_ex_cmd_mask; 2482 2483 char node_desc[IB_DEVICE_NODE_DESC_MAX]; 2484 __be64 node_guid; 2485 u32 local_dma_lkey; 2486 u16 is_switch:1; 2487 u8 node_type; 2488 u8 phys_port_cnt; 2489 struct ib_device_attr attrs; 2490 struct attribute_group *hw_stats_ag; 2491 struct rdma_hw_stats *hw_stats; 2492 2493 #ifdef CONFIG_CGROUP_RDMA 2494 struct rdmacg_device cg_device; 2495 #endif 2496 2497 u32 index; 2498 /* 2499 * Implementation details of the RDMA core, don't use in drivers 2500 */ 2501 struct rdma_restrack_root res; 2502 2503 /** 2504 * The following mandatory functions are used only at device 2505 * registration. Keep functions such as these at the end of this 2506 * structure to avoid cache line misses when accessing struct ib_device 2507 * in fast paths. 2508 */ 2509 int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *); 2510 void (*get_dev_fw_str)(struct ib_device *, char *str); 2511 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev, 2512 int comp_vector); 2513 2514 struct uverbs_root_spec *specs_root; 2515 enum rdma_driver_id driver_id; 2516 }; 2517 2518 struct ib_client { 2519 char *name; 2520 void (*add) (struct ib_device *); 2521 void (*remove)(struct ib_device *, void *client_data); 2522 2523 /* Returns the net_dev belonging to this ib_client and matching the 2524 * given parameters. 2525 * @dev: An RDMA device that the net_dev use for communication. 2526 * @port: A physical port number on the RDMA device. 2527 * @pkey: P_Key that the net_dev uses if applicable. 2528 * @gid: A GID that the net_dev uses to communicate. 2529 * @addr: An IP address the net_dev is configured with. 2530 * @client_data: The device's client data set by ib_set_client_data(). 2531 * 2532 * An ib_client that implements a net_dev on top of RDMA devices 2533 * (such as IP over IB) should implement this callback, allowing the 2534 * rdma_cm module to find the right net_dev for a given request. 2535 * 2536 * The caller is responsible for calling dev_put on the returned 2537 * netdev. */ 2538 struct net_device *(*get_net_dev_by_params)( 2539 struct ib_device *dev, 2540 u8 port, 2541 u16 pkey, 2542 const union ib_gid *gid, 2543 const struct sockaddr *addr, 2544 void *client_data); 2545 struct list_head list; 2546 }; 2547 2548 struct ib_device *ib_alloc_device(size_t size); 2549 void ib_dealloc_device(struct ib_device *device); 2550 2551 void ib_get_device_fw_str(struct ib_device *device, char *str); 2552 2553 int ib_register_device(struct ib_device *device, 2554 int (*port_callback)(struct ib_device *, 2555 u8, struct kobject *)); 2556 void ib_unregister_device(struct ib_device *device); 2557 2558 int ib_register_client (struct ib_client *client); 2559 void ib_unregister_client(struct ib_client *client); 2560 2561 void *ib_get_client_data(struct ib_device *device, struct ib_client *client); 2562 void ib_set_client_data(struct ib_device *device, struct ib_client *client, 2563 void *data); 2564 2565 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len) 2566 { 2567 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0; 2568 } 2569 2570 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len) 2571 { 2572 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0; 2573 } 2574 2575 static inline bool ib_is_buffer_cleared(const void __user *p, 2576 size_t len) 2577 { 2578 bool ret; 2579 u8 *buf; 2580 2581 if (len > USHRT_MAX) 2582 return false; 2583 2584 buf = memdup_user(p, len); 2585 if (IS_ERR(buf)) 2586 return false; 2587 2588 ret = !memchr_inv(buf, 0, len); 2589 kfree(buf); 2590 return ret; 2591 } 2592 2593 static inline bool ib_is_udata_cleared(struct ib_udata *udata, 2594 size_t offset, 2595 size_t len) 2596 { 2597 return ib_is_buffer_cleared(udata->inbuf + offset, len); 2598 } 2599 2600 /** 2601 * ib_modify_qp_is_ok - Check that the supplied attribute mask 2602 * contains all required attributes and no attributes not allowed for 2603 * the given QP state transition. 2604 * @cur_state: Current QP state 2605 * @next_state: Next QP state 2606 * @type: QP type 2607 * @mask: Mask of supplied QP attributes 2608 * @ll : link layer of port 2609 * 2610 * This function is a helper function that a low-level driver's 2611 * modify_qp method can use to validate the consumer's input. It 2612 * checks that cur_state and next_state are valid QP states, that a 2613 * transition from cur_state to next_state is allowed by the IB spec, 2614 * and that the attribute mask supplied is allowed for the transition. 2615 */ 2616 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state, 2617 enum ib_qp_type type, enum ib_qp_attr_mask mask, 2618 enum rdma_link_layer ll); 2619 2620 void ib_register_event_handler(struct ib_event_handler *event_handler); 2621 void ib_unregister_event_handler(struct ib_event_handler *event_handler); 2622 void ib_dispatch_event(struct ib_event *event); 2623 2624 int ib_query_port(struct ib_device *device, 2625 u8 port_num, struct ib_port_attr *port_attr); 2626 2627 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device, 2628 u8 port_num); 2629 2630 /** 2631 * rdma_cap_ib_switch - Check if the device is IB switch 2632 * @device: Device to check 2633 * 2634 * Device driver is responsible for setting is_switch bit on 2635 * in ib_device structure at init time. 2636 * 2637 * Return: true if the device is IB switch. 2638 */ 2639 static inline bool rdma_cap_ib_switch(const struct ib_device *device) 2640 { 2641 return device->is_switch; 2642 } 2643 2644 /** 2645 * rdma_start_port - Return the first valid port number for the device 2646 * specified 2647 * 2648 * @device: Device to be checked 2649 * 2650 * Return start port number 2651 */ 2652 static inline u8 rdma_start_port(const struct ib_device *device) 2653 { 2654 return rdma_cap_ib_switch(device) ? 0 : 1; 2655 } 2656 2657 /** 2658 * rdma_end_port - Return the last valid port number for the device 2659 * specified 2660 * 2661 * @device: Device to be checked 2662 * 2663 * Return last port number 2664 */ 2665 static inline u8 rdma_end_port(const struct ib_device *device) 2666 { 2667 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt; 2668 } 2669 2670 static inline int rdma_is_port_valid(const struct ib_device *device, 2671 unsigned int port) 2672 { 2673 return (port >= rdma_start_port(device) && 2674 port <= rdma_end_port(device)); 2675 } 2676 2677 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num) 2678 { 2679 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB; 2680 } 2681 2682 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num) 2683 { 2684 return device->port_immutable[port_num].core_cap_flags & 2685 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP); 2686 } 2687 2688 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num) 2689 { 2690 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP; 2691 } 2692 2693 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num) 2694 { 2695 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE; 2696 } 2697 2698 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num) 2699 { 2700 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP; 2701 } 2702 2703 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num) 2704 { 2705 return rdma_protocol_ib(device, port_num) || 2706 rdma_protocol_roce(device, port_num); 2707 } 2708 2709 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num) 2710 { 2711 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_RAW_PACKET; 2712 } 2713 2714 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num) 2715 { 2716 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_USNIC; 2717 } 2718 2719 /** 2720 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband 2721 * Management Datagrams. 2722 * @device: Device to check 2723 * @port_num: Port number to check 2724 * 2725 * Management Datagrams (MAD) are a required part of the InfiniBand 2726 * specification and are supported on all InfiniBand devices. A slightly 2727 * extended version are also supported on OPA interfaces. 2728 * 2729 * Return: true if the port supports sending/receiving of MAD packets. 2730 */ 2731 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num) 2732 { 2733 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD; 2734 } 2735 2736 /** 2737 * rdma_cap_opa_mad - Check if the port of device provides support for OPA 2738 * Management Datagrams. 2739 * @device: Device to check 2740 * @port_num: Port number to check 2741 * 2742 * Intel OmniPath devices extend and/or replace the InfiniBand Management 2743 * datagrams with their own versions. These OPA MADs share many but not all of 2744 * the characteristics of InfiniBand MADs. 2745 * 2746 * OPA MADs differ in the following ways: 2747 * 2748 * 1) MADs are variable size up to 2K 2749 * IBTA defined MADs remain fixed at 256 bytes 2750 * 2) OPA SMPs must carry valid PKeys 2751 * 3) OPA SMP packets are a different format 2752 * 2753 * Return: true if the port supports OPA MAD packet formats. 2754 */ 2755 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num) 2756 { 2757 return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD) 2758 == RDMA_CORE_CAP_OPA_MAD; 2759 } 2760 2761 /** 2762 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband 2763 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI). 2764 * @device: Device to check 2765 * @port_num: Port number to check 2766 * 2767 * Each InfiniBand node is required to provide a Subnet Management Agent 2768 * that the subnet manager can access. Prior to the fabric being fully 2769 * configured by the subnet manager, the SMA is accessed via a well known 2770 * interface called the Subnet Management Interface (SMI). This interface 2771 * uses directed route packets to communicate with the SM to get around the 2772 * chicken and egg problem of the SM needing to know what's on the fabric 2773 * in order to configure the fabric, and needing to configure the fabric in 2774 * order to send packets to the devices on the fabric. These directed 2775 * route packets do not need the fabric fully configured in order to reach 2776 * their destination. The SMI is the only method allowed to send 2777 * directed route packets on an InfiniBand fabric. 2778 * 2779 * Return: true if the port provides an SMI. 2780 */ 2781 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num) 2782 { 2783 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI; 2784 } 2785 2786 /** 2787 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband 2788 * Communication Manager. 2789 * @device: Device to check 2790 * @port_num: Port number to check 2791 * 2792 * The InfiniBand Communication Manager is one of many pre-defined General 2793 * Service Agents (GSA) that are accessed via the General Service 2794 * Interface (GSI). It's role is to facilitate establishment of connections 2795 * between nodes as well as other management related tasks for established 2796 * connections. 2797 * 2798 * Return: true if the port supports an IB CM (this does not guarantee that 2799 * a CM is actually running however). 2800 */ 2801 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num) 2802 { 2803 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM; 2804 } 2805 2806 /** 2807 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP 2808 * Communication Manager. 2809 * @device: Device to check 2810 * @port_num: Port number to check 2811 * 2812 * Similar to above, but specific to iWARP connections which have a different 2813 * managment protocol than InfiniBand. 2814 * 2815 * Return: true if the port supports an iWARP CM (this does not guarantee that 2816 * a CM is actually running however). 2817 */ 2818 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num) 2819 { 2820 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM; 2821 } 2822 2823 /** 2824 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband 2825 * Subnet Administration. 2826 * @device: Device to check 2827 * @port_num: Port number to check 2828 * 2829 * An InfiniBand Subnet Administration (SA) service is a pre-defined General 2830 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand 2831 * fabrics, devices should resolve routes to other hosts by contacting the 2832 * SA to query the proper route. 2833 * 2834 * Return: true if the port should act as a client to the fabric Subnet 2835 * Administration interface. This does not imply that the SA service is 2836 * running locally. 2837 */ 2838 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num) 2839 { 2840 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA; 2841 } 2842 2843 /** 2844 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband 2845 * Multicast. 2846 * @device: Device to check 2847 * @port_num: Port number to check 2848 * 2849 * InfiniBand multicast registration is more complex than normal IPv4 or 2850 * IPv6 multicast registration. Each Host Channel Adapter must register 2851 * with the Subnet Manager when it wishes to join a multicast group. It 2852 * should do so only once regardless of how many queue pairs it subscribes 2853 * to this group. And it should leave the group only after all queue pairs 2854 * attached to the group have been detached. 2855 * 2856 * Return: true if the port must undertake the additional adminstrative 2857 * overhead of registering/unregistering with the SM and tracking of the 2858 * total number of queue pairs attached to the multicast group. 2859 */ 2860 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num) 2861 { 2862 return rdma_cap_ib_sa(device, port_num); 2863 } 2864 2865 /** 2866 * rdma_cap_af_ib - Check if the port of device has the capability 2867 * Native Infiniband Address. 2868 * @device: Device to check 2869 * @port_num: Port number to check 2870 * 2871 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default 2872 * GID. RoCE uses a different mechanism, but still generates a GID via 2873 * a prescribed mechanism and port specific data. 2874 * 2875 * Return: true if the port uses a GID address to identify devices on the 2876 * network. 2877 */ 2878 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num) 2879 { 2880 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB; 2881 } 2882 2883 /** 2884 * rdma_cap_eth_ah - Check if the port of device has the capability 2885 * Ethernet Address Handle. 2886 * @device: Device to check 2887 * @port_num: Port number to check 2888 * 2889 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique 2890 * to fabricate GIDs over Ethernet/IP specific addresses native to the 2891 * port. Normally, packet headers are generated by the sending host 2892 * adapter, but when sending connectionless datagrams, we must manually 2893 * inject the proper headers for the fabric we are communicating over. 2894 * 2895 * Return: true if we are running as a RoCE port and must force the 2896 * addition of a Global Route Header built from our Ethernet Address 2897 * Handle into our header list for connectionless packets. 2898 */ 2899 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num) 2900 { 2901 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH; 2902 } 2903 2904 /** 2905 * rdma_cap_opa_ah - Check if the port of device supports 2906 * OPA Address handles 2907 * @device: Device to check 2908 * @port_num: Port number to check 2909 * 2910 * Return: true if we are running on an OPA device which supports 2911 * the extended OPA addressing. 2912 */ 2913 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num) 2914 { 2915 return (device->port_immutable[port_num].core_cap_flags & 2916 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH; 2917 } 2918 2919 /** 2920 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port. 2921 * 2922 * @device: Device 2923 * @port_num: Port number 2924 * 2925 * This MAD size includes the MAD headers and MAD payload. No other headers 2926 * are included. 2927 * 2928 * Return the max MAD size required by the Port. Will return 0 if the port 2929 * does not support MADs 2930 */ 2931 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num) 2932 { 2933 return device->port_immutable[port_num].max_mad_size; 2934 } 2935 2936 /** 2937 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table 2938 * @device: Device to check 2939 * @port_num: Port number to check 2940 * 2941 * RoCE GID table mechanism manages the various GIDs for a device. 2942 * 2943 * NOTE: if allocating the port's GID table has failed, this call will still 2944 * return true, but any RoCE GID table API will fail. 2945 * 2946 * Return: true if the port uses RoCE GID table mechanism in order to manage 2947 * its GIDs. 2948 */ 2949 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device, 2950 u8 port_num) 2951 { 2952 return rdma_protocol_roce(device, port_num) && 2953 device->add_gid && device->del_gid; 2954 } 2955 2956 /* 2957 * Check if the device supports READ W/ INVALIDATE. 2958 */ 2959 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num) 2960 { 2961 /* 2962 * iWarp drivers must support READ W/ INVALIDATE. No other protocol 2963 * has support for it yet. 2964 */ 2965 return rdma_protocol_iwarp(dev, port_num); 2966 } 2967 2968 int ib_query_gid(struct ib_device *device, 2969 u8 port_num, int index, union ib_gid *gid, 2970 struct ib_gid_attr *attr); 2971 2972 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port, 2973 int state); 2974 int ib_get_vf_config(struct ib_device *device, int vf, u8 port, 2975 struct ifla_vf_info *info); 2976 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port, 2977 struct ifla_vf_stats *stats); 2978 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid, 2979 int type); 2980 2981 int ib_query_pkey(struct ib_device *device, 2982 u8 port_num, u16 index, u16 *pkey); 2983 2984 int ib_modify_device(struct ib_device *device, 2985 int device_modify_mask, 2986 struct ib_device_modify *device_modify); 2987 2988 int ib_modify_port(struct ib_device *device, 2989 u8 port_num, int port_modify_mask, 2990 struct ib_port_modify *port_modify); 2991 2992 int ib_find_gid(struct ib_device *device, union ib_gid *gid, 2993 u8 *port_num, u16 *index); 2994 2995 int ib_find_pkey(struct ib_device *device, 2996 u8 port_num, u16 pkey, u16 *index); 2997 2998 enum ib_pd_flags { 2999 /* 3000 * Create a memory registration for all memory in the system and place 3001 * the rkey for it into pd->unsafe_global_rkey. This can be used by 3002 * ULPs to avoid the overhead of dynamic MRs. 3003 * 3004 * This flag is generally considered unsafe and must only be used in 3005 * extremly trusted environments. Every use of it will log a warning 3006 * in the kernel log. 3007 */ 3008 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01, 3009 }; 3010 3011 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags, 3012 const char *caller); 3013 #define ib_alloc_pd(device, flags) \ 3014 __ib_alloc_pd((device), (flags), KBUILD_MODNAME) 3015 void ib_dealloc_pd(struct ib_pd *pd); 3016 3017 /** 3018 * rdma_create_ah - Creates an address handle for the given address vector. 3019 * @pd: The protection domain associated with the address handle. 3020 * @ah_attr: The attributes of the address vector. 3021 * 3022 * The address handle is used to reference a local or global destination 3023 * in all UD QP post sends. 3024 */ 3025 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr); 3026 3027 /** 3028 * rdma_create_user_ah - Creates an address handle for the given address vector. 3029 * It resolves destination mac address for ah attribute of RoCE type. 3030 * @pd: The protection domain associated with the address handle. 3031 * @ah_attr: The attributes of the address vector. 3032 * @udata: pointer to user's input output buffer information need by 3033 * provider driver. 3034 * 3035 * It returns 0 on success and returns appropriate error code on error. 3036 * The address handle is used to reference a local or global destination 3037 * in all UD QP post sends. 3038 */ 3039 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd, 3040 struct rdma_ah_attr *ah_attr, 3041 struct ib_udata *udata); 3042 /** 3043 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header 3044 * work completion. 3045 * @hdr: the L3 header to parse 3046 * @net_type: type of header to parse 3047 * @sgid: place to store source gid 3048 * @dgid: place to store destination gid 3049 */ 3050 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr, 3051 enum rdma_network_type net_type, 3052 union ib_gid *sgid, union ib_gid *dgid); 3053 3054 /** 3055 * ib_get_rdma_header_version - Get the header version 3056 * @hdr: the L3 header to parse 3057 */ 3058 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr); 3059 3060 /** 3061 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a 3062 * work completion. 3063 * @device: Device on which the received message arrived. 3064 * @port_num: Port on which the received message arrived. 3065 * @wc: Work completion associated with the received message. 3066 * @grh: References the received global route header. This parameter is 3067 * ignored unless the work completion indicates that the GRH is valid. 3068 * @ah_attr: Returned attributes that can be used when creating an address 3069 * handle for replying to the message. 3070 */ 3071 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num, 3072 const struct ib_wc *wc, const struct ib_grh *grh, 3073 struct rdma_ah_attr *ah_attr); 3074 3075 /** 3076 * ib_create_ah_from_wc - Creates an address handle associated with the 3077 * sender of the specified work completion. 3078 * @pd: The protection domain associated with the address handle. 3079 * @wc: Work completion information associated with a received message. 3080 * @grh: References the received global route header. This parameter is 3081 * ignored unless the work completion indicates that the GRH is valid. 3082 * @port_num: The outbound port number to associate with the address. 3083 * 3084 * The address handle is used to reference a local or global destination 3085 * in all UD QP post sends. 3086 */ 3087 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc, 3088 const struct ib_grh *grh, u8 port_num); 3089 3090 /** 3091 * rdma_modify_ah - Modifies the address vector associated with an address 3092 * handle. 3093 * @ah: The address handle to modify. 3094 * @ah_attr: The new address vector attributes to associate with the 3095 * address handle. 3096 */ 3097 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr); 3098 3099 /** 3100 * rdma_query_ah - Queries the address vector associated with an address 3101 * handle. 3102 * @ah: The address handle to query. 3103 * @ah_attr: The address vector attributes associated with the address 3104 * handle. 3105 */ 3106 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr); 3107 3108 /** 3109 * rdma_destroy_ah - Destroys an address handle. 3110 * @ah: The address handle to destroy. 3111 */ 3112 int rdma_destroy_ah(struct ib_ah *ah); 3113 3114 /** 3115 * ib_create_srq - Creates a SRQ associated with the specified protection 3116 * domain. 3117 * @pd: The protection domain associated with the SRQ. 3118 * @srq_init_attr: A list of initial attributes required to create the 3119 * SRQ. If SRQ creation succeeds, then the attributes are updated to 3120 * the actual capabilities of the created SRQ. 3121 * 3122 * srq_attr->max_wr and srq_attr->max_sge are read the determine the 3123 * requested size of the SRQ, and set to the actual values allocated 3124 * on return. If ib_create_srq() succeeds, then max_wr and max_sge 3125 * will always be at least as large as the requested values. 3126 */ 3127 struct ib_srq *ib_create_srq(struct ib_pd *pd, 3128 struct ib_srq_init_attr *srq_init_attr); 3129 3130 /** 3131 * ib_modify_srq - Modifies the attributes for the specified SRQ. 3132 * @srq: The SRQ to modify. 3133 * @srq_attr: On input, specifies the SRQ attributes to modify. On output, 3134 * the current values of selected SRQ attributes are returned. 3135 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ 3136 * are being modified. 3137 * 3138 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or 3139 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when 3140 * the number of receives queued drops below the limit. 3141 */ 3142 int ib_modify_srq(struct ib_srq *srq, 3143 struct ib_srq_attr *srq_attr, 3144 enum ib_srq_attr_mask srq_attr_mask); 3145 3146 /** 3147 * ib_query_srq - Returns the attribute list and current values for the 3148 * specified SRQ. 3149 * @srq: The SRQ to query. 3150 * @srq_attr: The attributes of the specified SRQ. 3151 */ 3152 int ib_query_srq(struct ib_srq *srq, 3153 struct ib_srq_attr *srq_attr); 3154 3155 /** 3156 * ib_destroy_srq - Destroys the specified SRQ. 3157 * @srq: The SRQ to destroy. 3158 */ 3159 int ib_destroy_srq(struct ib_srq *srq); 3160 3161 /** 3162 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ. 3163 * @srq: The SRQ to post the work request on. 3164 * @recv_wr: A list of work requests to post on the receive queue. 3165 * @bad_recv_wr: On an immediate failure, this parameter will reference 3166 * the work request that failed to be posted on the QP. 3167 */ 3168 static inline int ib_post_srq_recv(struct ib_srq *srq, 3169 struct ib_recv_wr *recv_wr, 3170 struct ib_recv_wr **bad_recv_wr) 3171 { 3172 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr); 3173 } 3174 3175 /** 3176 * ib_create_qp - Creates a QP associated with the specified protection 3177 * domain. 3178 * @pd: The protection domain associated with the QP. 3179 * @qp_init_attr: A list of initial attributes required to create the 3180 * QP. If QP creation succeeds, then the attributes are updated to 3181 * the actual capabilities of the created QP. 3182 */ 3183 struct ib_qp *ib_create_qp(struct ib_pd *pd, 3184 struct ib_qp_init_attr *qp_init_attr); 3185 3186 /** 3187 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP. 3188 * @qp: The QP to modify. 3189 * @attr: On input, specifies the QP attributes to modify. On output, 3190 * the current values of selected QP attributes are returned. 3191 * @attr_mask: A bit-mask used to specify which attributes of the QP 3192 * are being modified. 3193 * @udata: pointer to user's input output buffer information 3194 * are being modified. 3195 * It returns 0 on success and returns appropriate error code on error. 3196 */ 3197 int ib_modify_qp_with_udata(struct ib_qp *qp, 3198 struct ib_qp_attr *attr, 3199 int attr_mask, 3200 struct ib_udata *udata); 3201 3202 /** 3203 * ib_modify_qp - Modifies the attributes for the specified QP and then 3204 * transitions the QP to the given state. 3205 * @qp: The QP to modify. 3206 * @qp_attr: On input, specifies the QP attributes to modify. On output, 3207 * the current values of selected QP attributes are returned. 3208 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP 3209 * are being modified. 3210 */ 3211 int ib_modify_qp(struct ib_qp *qp, 3212 struct ib_qp_attr *qp_attr, 3213 int qp_attr_mask); 3214 3215 /** 3216 * ib_query_qp - Returns the attribute list and current values for the 3217 * specified QP. 3218 * @qp: The QP to query. 3219 * @qp_attr: The attributes of the specified QP. 3220 * @qp_attr_mask: A bit-mask used to select specific attributes to query. 3221 * @qp_init_attr: Additional attributes of the selected QP. 3222 * 3223 * The qp_attr_mask may be used to limit the query to gathering only the 3224 * selected attributes. 3225 */ 3226 int ib_query_qp(struct ib_qp *qp, 3227 struct ib_qp_attr *qp_attr, 3228 int qp_attr_mask, 3229 struct ib_qp_init_attr *qp_init_attr); 3230 3231 /** 3232 * ib_destroy_qp - Destroys the specified QP. 3233 * @qp: The QP to destroy. 3234 */ 3235 int ib_destroy_qp(struct ib_qp *qp); 3236 3237 /** 3238 * ib_open_qp - Obtain a reference to an existing sharable QP. 3239 * @xrcd - XRC domain 3240 * @qp_open_attr: Attributes identifying the QP to open. 3241 * 3242 * Returns a reference to a sharable QP. 3243 */ 3244 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd, 3245 struct ib_qp_open_attr *qp_open_attr); 3246 3247 /** 3248 * ib_close_qp - Release an external reference to a QP. 3249 * @qp: The QP handle to release 3250 * 3251 * The opened QP handle is released by the caller. The underlying 3252 * shared QP is not destroyed until all internal references are released. 3253 */ 3254 int ib_close_qp(struct ib_qp *qp); 3255 3256 /** 3257 * ib_post_send - Posts a list of work requests to the send queue of 3258 * the specified QP. 3259 * @qp: The QP to post the work request on. 3260 * @send_wr: A list of work requests to post on the send queue. 3261 * @bad_send_wr: On an immediate failure, this parameter will reference 3262 * the work request that failed to be posted on the QP. 3263 * 3264 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate 3265 * error is returned, the QP state shall not be affected, 3266 * ib_post_send() will return an immediate error after queueing any 3267 * earlier work requests in the list. 3268 */ 3269 static inline int ib_post_send(struct ib_qp *qp, 3270 struct ib_send_wr *send_wr, 3271 struct ib_send_wr **bad_send_wr) 3272 { 3273 return qp->device->post_send(qp, send_wr, bad_send_wr); 3274 } 3275 3276 /** 3277 * ib_post_recv - Posts a list of work requests to the receive queue of 3278 * the specified QP. 3279 * @qp: The QP to post the work request on. 3280 * @recv_wr: A list of work requests to post on the receive queue. 3281 * @bad_recv_wr: On an immediate failure, this parameter will reference 3282 * the work request that failed to be posted on the QP. 3283 */ 3284 static inline int ib_post_recv(struct ib_qp *qp, 3285 struct ib_recv_wr *recv_wr, 3286 struct ib_recv_wr **bad_recv_wr) 3287 { 3288 return qp->device->post_recv(qp, recv_wr, bad_recv_wr); 3289 } 3290 3291 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private, 3292 int nr_cqe, int comp_vector, 3293 enum ib_poll_context poll_ctx, const char *caller); 3294 #define ib_alloc_cq(device, priv, nr_cqe, comp_vect, poll_ctx) \ 3295 __ib_alloc_cq((device), (priv), (nr_cqe), (comp_vect), (poll_ctx), KBUILD_MODNAME) 3296 3297 void ib_free_cq(struct ib_cq *cq); 3298 int ib_process_cq_direct(struct ib_cq *cq, int budget); 3299 3300 /** 3301 * ib_create_cq - Creates a CQ on the specified device. 3302 * @device: The device on which to create the CQ. 3303 * @comp_handler: A user-specified callback that is invoked when a 3304 * completion event occurs on the CQ. 3305 * @event_handler: A user-specified callback that is invoked when an 3306 * asynchronous event not associated with a completion occurs on the CQ. 3307 * @cq_context: Context associated with the CQ returned to the user via 3308 * the associated completion and event handlers. 3309 * @cq_attr: The attributes the CQ should be created upon. 3310 * 3311 * Users can examine the cq structure to determine the actual CQ size. 3312 */ 3313 struct ib_cq *ib_create_cq(struct ib_device *device, 3314 ib_comp_handler comp_handler, 3315 void (*event_handler)(struct ib_event *, void *), 3316 void *cq_context, 3317 const struct ib_cq_init_attr *cq_attr); 3318 3319 /** 3320 * ib_resize_cq - Modifies the capacity of the CQ. 3321 * @cq: The CQ to resize. 3322 * @cqe: The minimum size of the CQ. 3323 * 3324 * Users can examine the cq structure to determine the actual CQ size. 3325 */ 3326 int ib_resize_cq(struct ib_cq *cq, int cqe); 3327 3328 /** 3329 * rdma_set_cq_moderation - Modifies moderation params of the CQ 3330 * @cq: The CQ to modify. 3331 * @cq_count: number of CQEs that will trigger an event 3332 * @cq_period: max period of time in usec before triggering an event 3333 * 3334 */ 3335 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period); 3336 3337 /** 3338 * ib_destroy_cq - Destroys the specified CQ. 3339 * @cq: The CQ to destroy. 3340 */ 3341 int ib_destroy_cq(struct ib_cq *cq); 3342 3343 /** 3344 * ib_poll_cq - poll a CQ for completion(s) 3345 * @cq:the CQ being polled 3346 * @num_entries:maximum number of completions to return 3347 * @wc:array of at least @num_entries &struct ib_wc where completions 3348 * will be returned 3349 * 3350 * Poll a CQ for (possibly multiple) completions. If the return value 3351 * is < 0, an error occurred. If the return value is >= 0, it is the 3352 * number of completions returned. If the return value is 3353 * non-negative and < num_entries, then the CQ was emptied. 3354 */ 3355 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries, 3356 struct ib_wc *wc) 3357 { 3358 return cq->device->poll_cq(cq, num_entries, wc); 3359 } 3360 3361 /** 3362 * ib_req_notify_cq - Request completion notification on a CQ. 3363 * @cq: The CQ to generate an event for. 3364 * @flags: 3365 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP 3366 * to request an event on the next solicited event or next work 3367 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS 3368 * may also be |ed in to request a hint about missed events, as 3369 * described below. 3370 * 3371 * Return Value: 3372 * < 0 means an error occurred while requesting notification 3373 * == 0 means notification was requested successfully, and if 3374 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events 3375 * were missed and it is safe to wait for another event. In 3376 * this case is it guaranteed that any work completions added 3377 * to the CQ since the last CQ poll will trigger a completion 3378 * notification event. 3379 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed 3380 * in. It means that the consumer must poll the CQ again to 3381 * make sure it is empty to avoid missing an event because of a 3382 * race between requesting notification and an entry being 3383 * added to the CQ. This return value means it is possible 3384 * (but not guaranteed) that a work completion has been added 3385 * to the CQ since the last poll without triggering a 3386 * completion notification event. 3387 */ 3388 static inline int ib_req_notify_cq(struct ib_cq *cq, 3389 enum ib_cq_notify_flags flags) 3390 { 3391 return cq->device->req_notify_cq(cq, flags); 3392 } 3393 3394 /** 3395 * ib_req_ncomp_notif - Request completion notification when there are 3396 * at least the specified number of unreaped completions on the CQ. 3397 * @cq: The CQ to generate an event for. 3398 * @wc_cnt: The number of unreaped completions that should be on the 3399 * CQ before an event is generated. 3400 */ 3401 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt) 3402 { 3403 return cq->device->req_ncomp_notif ? 3404 cq->device->req_ncomp_notif(cq, wc_cnt) : 3405 -ENOSYS; 3406 } 3407 3408 /** 3409 * ib_dma_mapping_error - check a DMA addr for error 3410 * @dev: The device for which the dma_addr was created 3411 * @dma_addr: The DMA address to check 3412 */ 3413 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr) 3414 { 3415 return dma_mapping_error(dev->dma_device, dma_addr); 3416 } 3417 3418 /** 3419 * ib_dma_map_single - Map a kernel virtual address to DMA address 3420 * @dev: The device for which the dma_addr is to be created 3421 * @cpu_addr: The kernel virtual address 3422 * @size: The size of the region in bytes 3423 * @direction: The direction of the DMA 3424 */ 3425 static inline u64 ib_dma_map_single(struct ib_device *dev, 3426 void *cpu_addr, size_t size, 3427 enum dma_data_direction direction) 3428 { 3429 return dma_map_single(dev->dma_device, cpu_addr, size, direction); 3430 } 3431 3432 /** 3433 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single() 3434 * @dev: The device for which the DMA address was created 3435 * @addr: The DMA address 3436 * @size: The size of the region in bytes 3437 * @direction: The direction of the DMA 3438 */ 3439 static inline void ib_dma_unmap_single(struct ib_device *dev, 3440 u64 addr, size_t size, 3441 enum dma_data_direction direction) 3442 { 3443 dma_unmap_single(dev->dma_device, addr, size, direction); 3444 } 3445 3446 /** 3447 * ib_dma_map_page - Map a physical page to DMA address 3448 * @dev: The device for which the dma_addr is to be created 3449 * @page: The page to be mapped 3450 * @offset: The offset within the page 3451 * @size: The size of the region in bytes 3452 * @direction: The direction of the DMA 3453 */ 3454 static inline u64 ib_dma_map_page(struct ib_device *dev, 3455 struct page *page, 3456 unsigned long offset, 3457 size_t size, 3458 enum dma_data_direction direction) 3459 { 3460 return dma_map_page(dev->dma_device, page, offset, size, direction); 3461 } 3462 3463 /** 3464 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page() 3465 * @dev: The device for which the DMA address was created 3466 * @addr: The DMA address 3467 * @size: The size of the region in bytes 3468 * @direction: The direction of the DMA 3469 */ 3470 static inline void ib_dma_unmap_page(struct ib_device *dev, 3471 u64 addr, size_t size, 3472 enum dma_data_direction direction) 3473 { 3474 dma_unmap_page(dev->dma_device, addr, size, direction); 3475 } 3476 3477 /** 3478 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses 3479 * @dev: The device for which the DMA addresses are to be created 3480 * @sg: The array of scatter/gather entries 3481 * @nents: The number of scatter/gather entries 3482 * @direction: The direction of the DMA 3483 */ 3484 static inline int ib_dma_map_sg(struct ib_device *dev, 3485 struct scatterlist *sg, int nents, 3486 enum dma_data_direction direction) 3487 { 3488 return dma_map_sg(dev->dma_device, sg, nents, direction); 3489 } 3490 3491 /** 3492 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses 3493 * @dev: The device for which the DMA addresses were created 3494 * @sg: The array of scatter/gather entries 3495 * @nents: The number of scatter/gather entries 3496 * @direction: The direction of the DMA 3497 */ 3498 static inline void ib_dma_unmap_sg(struct ib_device *dev, 3499 struct scatterlist *sg, int nents, 3500 enum dma_data_direction direction) 3501 { 3502 dma_unmap_sg(dev->dma_device, sg, nents, direction); 3503 } 3504 3505 static inline int ib_dma_map_sg_attrs(struct ib_device *dev, 3506 struct scatterlist *sg, int nents, 3507 enum dma_data_direction direction, 3508 unsigned long dma_attrs) 3509 { 3510 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, 3511 dma_attrs); 3512 } 3513 3514 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev, 3515 struct scatterlist *sg, int nents, 3516 enum dma_data_direction direction, 3517 unsigned long dma_attrs) 3518 { 3519 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs); 3520 } 3521 /** 3522 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry 3523 * @dev: The device for which the DMA addresses were created 3524 * @sg: The scatter/gather entry 3525 * 3526 * Note: this function is obsolete. To do: change all occurrences of 3527 * ib_sg_dma_address() into sg_dma_address(). 3528 */ 3529 static inline u64 ib_sg_dma_address(struct ib_device *dev, 3530 struct scatterlist *sg) 3531 { 3532 return sg_dma_address(sg); 3533 } 3534 3535 /** 3536 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry 3537 * @dev: The device for which the DMA addresses were created 3538 * @sg: The scatter/gather entry 3539 * 3540 * Note: this function is obsolete. To do: change all occurrences of 3541 * ib_sg_dma_len() into sg_dma_len(). 3542 */ 3543 static inline unsigned int ib_sg_dma_len(struct ib_device *dev, 3544 struct scatterlist *sg) 3545 { 3546 return sg_dma_len(sg); 3547 } 3548 3549 /** 3550 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU 3551 * @dev: The device for which the DMA address was created 3552 * @addr: The DMA address 3553 * @size: The size of the region in bytes 3554 * @dir: The direction of the DMA 3555 */ 3556 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev, 3557 u64 addr, 3558 size_t size, 3559 enum dma_data_direction dir) 3560 { 3561 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir); 3562 } 3563 3564 /** 3565 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device 3566 * @dev: The device for which the DMA address was created 3567 * @addr: The DMA address 3568 * @size: The size of the region in bytes 3569 * @dir: The direction of the DMA 3570 */ 3571 static inline void ib_dma_sync_single_for_device(struct ib_device *dev, 3572 u64 addr, 3573 size_t size, 3574 enum dma_data_direction dir) 3575 { 3576 dma_sync_single_for_device(dev->dma_device, addr, size, dir); 3577 } 3578 3579 /** 3580 * ib_dma_alloc_coherent - Allocate memory and map it for DMA 3581 * @dev: The device for which the DMA address is requested 3582 * @size: The size of the region to allocate in bytes 3583 * @dma_handle: A pointer for returning the DMA address of the region 3584 * @flag: memory allocator flags 3585 */ 3586 static inline void *ib_dma_alloc_coherent(struct ib_device *dev, 3587 size_t size, 3588 dma_addr_t *dma_handle, 3589 gfp_t flag) 3590 { 3591 return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag); 3592 } 3593 3594 /** 3595 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent() 3596 * @dev: The device for which the DMA addresses were allocated 3597 * @size: The size of the region 3598 * @cpu_addr: the address returned by ib_dma_alloc_coherent() 3599 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent() 3600 */ 3601 static inline void ib_dma_free_coherent(struct ib_device *dev, 3602 size_t size, void *cpu_addr, 3603 dma_addr_t dma_handle) 3604 { 3605 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle); 3606 } 3607 3608 /** 3609 * ib_dereg_mr - Deregisters a memory region and removes it from the 3610 * HCA translation table. 3611 * @mr: The memory region to deregister. 3612 * 3613 * This function can fail, if the memory region has memory windows bound to it. 3614 */ 3615 int ib_dereg_mr(struct ib_mr *mr); 3616 3617 struct ib_mr *ib_alloc_mr(struct ib_pd *pd, 3618 enum ib_mr_type mr_type, 3619 u32 max_num_sg); 3620 3621 /** 3622 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR 3623 * R_Key and L_Key. 3624 * @mr - struct ib_mr pointer to be updated. 3625 * @newkey - new key to be used. 3626 */ 3627 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey) 3628 { 3629 mr->lkey = (mr->lkey & 0xffffff00) | newkey; 3630 mr->rkey = (mr->rkey & 0xffffff00) | newkey; 3631 } 3632 3633 /** 3634 * ib_inc_rkey - increments the key portion of the given rkey. Can be used 3635 * for calculating a new rkey for type 2 memory windows. 3636 * @rkey - the rkey to increment. 3637 */ 3638 static inline u32 ib_inc_rkey(u32 rkey) 3639 { 3640 const u32 mask = 0x000000ff; 3641 return ((rkey + 1) & mask) | (rkey & ~mask); 3642 } 3643 3644 /** 3645 * ib_alloc_fmr - Allocates a unmapped fast memory region. 3646 * @pd: The protection domain associated with the unmapped region. 3647 * @mr_access_flags: Specifies the memory access rights. 3648 * @fmr_attr: Attributes of the unmapped region. 3649 * 3650 * A fast memory region must be mapped before it can be used as part of 3651 * a work request. 3652 */ 3653 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd, 3654 int mr_access_flags, 3655 struct ib_fmr_attr *fmr_attr); 3656 3657 /** 3658 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region. 3659 * @fmr: The fast memory region to associate with the pages. 3660 * @page_list: An array of physical pages to map to the fast memory region. 3661 * @list_len: The number of pages in page_list. 3662 * @iova: The I/O virtual address to use with the mapped region. 3663 */ 3664 static inline int ib_map_phys_fmr(struct ib_fmr *fmr, 3665 u64 *page_list, int list_len, 3666 u64 iova) 3667 { 3668 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova); 3669 } 3670 3671 /** 3672 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions. 3673 * @fmr_list: A linked list of fast memory regions to unmap. 3674 */ 3675 int ib_unmap_fmr(struct list_head *fmr_list); 3676 3677 /** 3678 * ib_dealloc_fmr - Deallocates a fast memory region. 3679 * @fmr: The fast memory region to deallocate. 3680 */ 3681 int ib_dealloc_fmr(struct ib_fmr *fmr); 3682 3683 /** 3684 * ib_attach_mcast - Attaches the specified QP to a multicast group. 3685 * @qp: QP to attach to the multicast group. The QP must be type 3686 * IB_QPT_UD. 3687 * @gid: Multicast group GID. 3688 * @lid: Multicast group LID in host byte order. 3689 * 3690 * In order to send and receive multicast packets, subnet 3691 * administration must have created the multicast group and configured 3692 * the fabric appropriately. The port associated with the specified 3693 * QP must also be a member of the multicast group. 3694 */ 3695 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid); 3696 3697 /** 3698 * ib_detach_mcast - Detaches the specified QP from a multicast group. 3699 * @qp: QP to detach from the multicast group. 3700 * @gid: Multicast group GID. 3701 * @lid: Multicast group LID in host byte order. 3702 */ 3703 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid); 3704 3705 /** 3706 * ib_alloc_xrcd - Allocates an XRC domain. 3707 * @device: The device on which to allocate the XRC domain. 3708 * @caller: Module name for kernel consumers 3709 */ 3710 struct ib_xrcd *__ib_alloc_xrcd(struct ib_device *device, const char *caller); 3711 #define ib_alloc_xrcd(device) \ 3712 __ib_alloc_xrcd((device), KBUILD_MODNAME) 3713 3714 /** 3715 * ib_dealloc_xrcd - Deallocates an XRC domain. 3716 * @xrcd: The XRC domain to deallocate. 3717 */ 3718 int ib_dealloc_xrcd(struct ib_xrcd *xrcd); 3719 3720 struct ib_flow *ib_create_flow(struct ib_qp *qp, 3721 struct ib_flow_attr *flow_attr, int domain); 3722 int ib_destroy_flow(struct ib_flow *flow_id); 3723 3724 static inline int ib_check_mr_access(int flags) 3725 { 3726 /* 3727 * Local write permission is required if remote write or 3728 * remote atomic permission is also requested. 3729 */ 3730 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) && 3731 !(flags & IB_ACCESS_LOCAL_WRITE)) 3732 return -EINVAL; 3733 3734 return 0; 3735 } 3736 3737 /** 3738 * ib_check_mr_status: lightweight check of MR status. 3739 * This routine may provide status checks on a selected 3740 * ib_mr. first use is for signature status check. 3741 * 3742 * @mr: A memory region. 3743 * @check_mask: Bitmask of which checks to perform from 3744 * ib_mr_status_check enumeration. 3745 * @mr_status: The container of relevant status checks. 3746 * failed checks will be indicated in the status bitmask 3747 * and the relevant info shall be in the error item. 3748 */ 3749 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask, 3750 struct ib_mr_status *mr_status); 3751 3752 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port, 3753 u16 pkey, const union ib_gid *gid, 3754 const struct sockaddr *addr); 3755 struct ib_wq *ib_create_wq(struct ib_pd *pd, 3756 struct ib_wq_init_attr *init_attr); 3757 int ib_destroy_wq(struct ib_wq *wq); 3758 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr, 3759 u32 wq_attr_mask); 3760 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device, 3761 struct ib_rwq_ind_table_init_attr* 3762 wq_ind_table_init_attr); 3763 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table); 3764 3765 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents, 3766 unsigned int *sg_offset, unsigned int page_size); 3767 3768 static inline int 3769 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents, 3770 unsigned int *sg_offset, unsigned int page_size) 3771 { 3772 int n; 3773 3774 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size); 3775 mr->iova = 0; 3776 3777 return n; 3778 } 3779 3780 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents, 3781 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64)); 3782 3783 void ib_drain_rq(struct ib_qp *qp); 3784 void ib_drain_sq(struct ib_qp *qp); 3785 void ib_drain_qp(struct ib_qp *qp); 3786 3787 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width); 3788 3789 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr) 3790 { 3791 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE) 3792 return attr->roce.dmac; 3793 return NULL; 3794 } 3795 3796 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid) 3797 { 3798 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 3799 attr->ib.dlid = (u16)dlid; 3800 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 3801 attr->opa.dlid = dlid; 3802 } 3803 3804 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr) 3805 { 3806 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 3807 return attr->ib.dlid; 3808 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 3809 return attr->opa.dlid; 3810 return 0; 3811 } 3812 3813 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl) 3814 { 3815 attr->sl = sl; 3816 } 3817 3818 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr) 3819 { 3820 return attr->sl; 3821 } 3822 3823 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr, 3824 u8 src_path_bits) 3825 { 3826 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 3827 attr->ib.src_path_bits = src_path_bits; 3828 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 3829 attr->opa.src_path_bits = src_path_bits; 3830 } 3831 3832 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr) 3833 { 3834 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 3835 return attr->ib.src_path_bits; 3836 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 3837 return attr->opa.src_path_bits; 3838 return 0; 3839 } 3840 3841 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr, 3842 bool make_grd) 3843 { 3844 if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 3845 attr->opa.make_grd = make_grd; 3846 } 3847 3848 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr) 3849 { 3850 if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 3851 return attr->opa.make_grd; 3852 return false; 3853 } 3854 3855 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num) 3856 { 3857 attr->port_num = port_num; 3858 } 3859 3860 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr) 3861 { 3862 return attr->port_num; 3863 } 3864 3865 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr, 3866 u8 static_rate) 3867 { 3868 attr->static_rate = static_rate; 3869 } 3870 3871 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr) 3872 { 3873 return attr->static_rate; 3874 } 3875 3876 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr, 3877 enum ib_ah_flags flag) 3878 { 3879 attr->ah_flags = flag; 3880 } 3881 3882 static inline enum ib_ah_flags 3883 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr) 3884 { 3885 return attr->ah_flags; 3886 } 3887 3888 static inline const struct ib_global_route 3889 *rdma_ah_read_grh(const struct rdma_ah_attr *attr) 3890 { 3891 return &attr->grh; 3892 } 3893 3894 /*To retrieve and modify the grh */ 3895 static inline struct ib_global_route 3896 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr) 3897 { 3898 return &attr->grh; 3899 } 3900 3901 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid) 3902 { 3903 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 3904 3905 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid)); 3906 } 3907 3908 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr, 3909 __be64 prefix) 3910 { 3911 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 3912 3913 grh->dgid.global.subnet_prefix = prefix; 3914 } 3915 3916 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr, 3917 __be64 if_id) 3918 { 3919 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 3920 3921 grh->dgid.global.interface_id = if_id; 3922 } 3923 3924 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr, 3925 union ib_gid *dgid, u32 flow_label, 3926 u8 sgid_index, u8 hop_limit, 3927 u8 traffic_class) 3928 { 3929 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 3930 3931 attr->ah_flags = IB_AH_GRH; 3932 if (dgid) 3933 grh->dgid = *dgid; 3934 grh->flow_label = flow_label; 3935 grh->sgid_index = sgid_index; 3936 grh->hop_limit = hop_limit; 3937 grh->traffic_class = traffic_class; 3938 } 3939 3940 /** 3941 * rdma_ah_find_type - Return address handle type. 3942 * 3943 * @dev: Device to be checked 3944 * @port_num: Port number 3945 */ 3946 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev, 3947 u8 port_num) 3948 { 3949 if (rdma_protocol_roce(dev, port_num)) 3950 return RDMA_AH_ATTR_TYPE_ROCE; 3951 if (rdma_protocol_ib(dev, port_num)) { 3952 if (rdma_cap_opa_ah(dev, port_num)) 3953 return RDMA_AH_ATTR_TYPE_OPA; 3954 return RDMA_AH_ATTR_TYPE_IB; 3955 } 3956 3957 return RDMA_AH_ATTR_TYPE_UNDEFINED; 3958 } 3959 3960 /** 3961 * ib_lid_cpu16 - Return lid in 16bit CPU encoding. 3962 * In the current implementation the only way to get 3963 * get the 32bit lid is from other sources for OPA. 3964 * For IB, lids will always be 16bits so cast the 3965 * value accordingly. 3966 * 3967 * @lid: A 32bit LID 3968 */ 3969 static inline u16 ib_lid_cpu16(u32 lid) 3970 { 3971 WARN_ON_ONCE(lid & 0xFFFF0000); 3972 return (u16)lid; 3973 } 3974 3975 /** 3976 * ib_lid_be16 - Return lid in 16bit BE encoding. 3977 * 3978 * @lid: A 32bit LID 3979 */ 3980 static inline __be16 ib_lid_be16(u32 lid) 3981 { 3982 WARN_ON_ONCE(lid & 0xFFFF0000); 3983 return cpu_to_be16((u16)lid); 3984 } 3985 3986 /** 3987 * ib_get_vector_affinity - Get the affinity mappings of a given completion 3988 * vector 3989 * @device: the rdma device 3990 * @comp_vector: index of completion vector 3991 * 3992 * Returns NULL on failure, otherwise a corresponding cpu map of the 3993 * completion vector (returns all-cpus map if the device driver doesn't 3994 * implement get_vector_affinity). 3995 */ 3996 static inline const struct cpumask * 3997 ib_get_vector_affinity(struct ib_device *device, int comp_vector) 3998 { 3999 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors || 4000 !device->get_vector_affinity) 4001 return NULL; 4002 4003 return device->get_vector_affinity(device, comp_vector); 4004 4005 } 4006 4007 /** 4008 * rdma_roce_rescan_device - Rescan all of the network devices in the system 4009 * and add their gids, as needed, to the relevant RoCE devices. 4010 * 4011 * @device: the rdma device 4012 */ 4013 void rdma_roce_rescan_device(struct ib_device *ibdev); 4014 4015 #endif /* IB_VERBS_H */ 4016