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 57 #include <linux/atomic.h> 58 #include <linux/mmu_notifier.h> 59 #include <asm/uaccess.h> 60 61 extern struct workqueue_struct *ib_wq; 62 extern struct workqueue_struct *ib_comp_wq; 63 64 union ib_gid { 65 u8 raw[16]; 66 struct { 67 __be64 subnet_prefix; 68 __be64 interface_id; 69 } global; 70 }; 71 72 extern union ib_gid zgid; 73 74 enum ib_gid_type { 75 /* If link layer is Ethernet, this is RoCE V1 */ 76 IB_GID_TYPE_IB = 0, 77 IB_GID_TYPE_ROCE = 0, 78 IB_GID_TYPE_ROCE_UDP_ENCAP = 1, 79 IB_GID_TYPE_SIZE 80 }; 81 82 struct ib_gid_attr { 83 enum ib_gid_type gid_type; 84 struct net_device *ndev; 85 }; 86 87 enum rdma_node_type { 88 /* IB values map to NodeInfo:NodeType. */ 89 RDMA_NODE_IB_CA = 1, 90 RDMA_NODE_IB_SWITCH, 91 RDMA_NODE_IB_ROUTER, 92 RDMA_NODE_RNIC, 93 RDMA_NODE_USNIC, 94 RDMA_NODE_USNIC_UDP, 95 }; 96 97 enum rdma_transport_type { 98 RDMA_TRANSPORT_IB, 99 RDMA_TRANSPORT_IWARP, 100 RDMA_TRANSPORT_USNIC, 101 RDMA_TRANSPORT_USNIC_UDP 102 }; 103 104 enum rdma_protocol_type { 105 RDMA_PROTOCOL_IB, 106 RDMA_PROTOCOL_IBOE, 107 RDMA_PROTOCOL_IWARP, 108 RDMA_PROTOCOL_USNIC_UDP 109 }; 110 111 __attribute_const__ enum rdma_transport_type 112 rdma_node_get_transport(enum rdma_node_type node_type); 113 114 enum rdma_network_type { 115 RDMA_NETWORK_IB, 116 RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB, 117 RDMA_NETWORK_IPV4, 118 RDMA_NETWORK_IPV6 119 }; 120 121 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type) 122 { 123 if (network_type == RDMA_NETWORK_IPV4 || 124 network_type == RDMA_NETWORK_IPV6) 125 return IB_GID_TYPE_ROCE_UDP_ENCAP; 126 127 /* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */ 128 return IB_GID_TYPE_IB; 129 } 130 131 static inline enum rdma_network_type ib_gid_to_network_type(enum ib_gid_type gid_type, 132 union ib_gid *gid) 133 { 134 if (gid_type == IB_GID_TYPE_IB) 135 return RDMA_NETWORK_IB; 136 137 if (ipv6_addr_v4mapped((struct in6_addr *)gid)) 138 return RDMA_NETWORK_IPV4; 139 else 140 return RDMA_NETWORK_IPV6; 141 } 142 143 enum rdma_link_layer { 144 IB_LINK_LAYER_UNSPECIFIED, 145 IB_LINK_LAYER_INFINIBAND, 146 IB_LINK_LAYER_ETHERNET, 147 }; 148 149 enum ib_device_cap_flags { 150 IB_DEVICE_RESIZE_MAX_WR = 1, 151 IB_DEVICE_BAD_PKEY_CNTR = (1<<1), 152 IB_DEVICE_BAD_QKEY_CNTR = (1<<2), 153 IB_DEVICE_RAW_MULTI = (1<<3), 154 IB_DEVICE_AUTO_PATH_MIG = (1<<4), 155 IB_DEVICE_CHANGE_PHY_PORT = (1<<5), 156 IB_DEVICE_UD_AV_PORT_ENFORCE = (1<<6), 157 IB_DEVICE_CURR_QP_STATE_MOD = (1<<7), 158 IB_DEVICE_SHUTDOWN_PORT = (1<<8), 159 IB_DEVICE_INIT_TYPE = (1<<9), 160 IB_DEVICE_PORT_ACTIVE_EVENT = (1<<10), 161 IB_DEVICE_SYS_IMAGE_GUID = (1<<11), 162 IB_DEVICE_RC_RNR_NAK_GEN = (1<<12), 163 IB_DEVICE_SRQ_RESIZE = (1<<13), 164 IB_DEVICE_N_NOTIFY_CQ = (1<<14), 165 166 /* 167 * This device supports a per-device lkey or stag that can be 168 * used without performing a memory registration for the local 169 * memory. Note that ULPs should never check this flag, but 170 * instead of use the local_dma_lkey flag in the ib_pd structure, 171 * which will always contain a usable lkey. 172 */ 173 IB_DEVICE_LOCAL_DMA_LKEY = (1<<15), 174 IB_DEVICE_RESERVED = (1<<16), /* old SEND_W_INV */ 175 IB_DEVICE_MEM_WINDOW = (1<<17), 176 /* 177 * Devices should set IB_DEVICE_UD_IP_SUM if they support 178 * insertion of UDP and TCP checksum on outgoing UD IPoIB 179 * messages and can verify the validity of checksum for 180 * incoming messages. Setting this flag implies that the 181 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode. 182 */ 183 IB_DEVICE_UD_IP_CSUM = (1<<18), 184 IB_DEVICE_UD_TSO = (1<<19), 185 IB_DEVICE_XRC = (1<<20), 186 187 /* 188 * This device supports the IB "base memory management extension", 189 * which includes support for fast registrations (IB_WR_REG_MR, 190 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should 191 * also be set by any iWarp device which must support FRs to comply 192 * to the iWarp verbs spec. iWarp devices also support the 193 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the 194 * stag. 195 */ 196 IB_DEVICE_MEM_MGT_EXTENSIONS = (1<<21), 197 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1<<22), 198 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1<<23), 199 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1<<24), 200 IB_DEVICE_RC_IP_CSUM = (1<<25), 201 IB_DEVICE_RAW_IP_CSUM = (1<<26), 202 IB_DEVICE_MANAGED_FLOW_STEERING = (1<<29), 203 IB_DEVICE_SIGNATURE_HANDOVER = (1<<30), 204 IB_DEVICE_ON_DEMAND_PAGING = (1<<31), 205 }; 206 207 enum ib_signature_prot_cap { 208 IB_PROT_T10DIF_TYPE_1 = 1, 209 IB_PROT_T10DIF_TYPE_2 = 1 << 1, 210 IB_PROT_T10DIF_TYPE_3 = 1 << 2, 211 }; 212 213 enum ib_signature_guard_cap { 214 IB_GUARD_T10DIF_CRC = 1, 215 IB_GUARD_T10DIF_CSUM = 1 << 1, 216 }; 217 218 enum ib_atomic_cap { 219 IB_ATOMIC_NONE, 220 IB_ATOMIC_HCA, 221 IB_ATOMIC_GLOB 222 }; 223 224 enum ib_odp_general_cap_bits { 225 IB_ODP_SUPPORT = 1 << 0, 226 }; 227 228 enum ib_odp_transport_cap_bits { 229 IB_ODP_SUPPORT_SEND = 1 << 0, 230 IB_ODP_SUPPORT_RECV = 1 << 1, 231 IB_ODP_SUPPORT_WRITE = 1 << 2, 232 IB_ODP_SUPPORT_READ = 1 << 3, 233 IB_ODP_SUPPORT_ATOMIC = 1 << 4, 234 }; 235 236 struct ib_odp_caps { 237 uint64_t general_caps; 238 struct { 239 uint32_t rc_odp_caps; 240 uint32_t uc_odp_caps; 241 uint32_t ud_odp_caps; 242 } per_transport_caps; 243 }; 244 245 enum ib_cq_creation_flags { 246 IB_CQ_FLAGS_TIMESTAMP_COMPLETION = 1 << 0, 247 }; 248 249 struct ib_cq_init_attr { 250 unsigned int cqe; 251 int comp_vector; 252 u32 flags; 253 }; 254 255 struct ib_device_attr { 256 u64 fw_ver; 257 __be64 sys_image_guid; 258 u64 max_mr_size; 259 u64 page_size_cap; 260 u32 vendor_id; 261 u32 vendor_part_id; 262 u32 hw_ver; 263 int max_qp; 264 int max_qp_wr; 265 int device_cap_flags; 266 int max_sge; 267 int max_sge_rd; 268 int max_cq; 269 int max_cqe; 270 int max_mr; 271 int max_pd; 272 int max_qp_rd_atom; 273 int max_ee_rd_atom; 274 int max_res_rd_atom; 275 int max_qp_init_rd_atom; 276 int max_ee_init_rd_atom; 277 enum ib_atomic_cap atomic_cap; 278 enum ib_atomic_cap masked_atomic_cap; 279 int max_ee; 280 int max_rdd; 281 int max_mw; 282 int max_raw_ipv6_qp; 283 int max_raw_ethy_qp; 284 int max_mcast_grp; 285 int max_mcast_qp_attach; 286 int max_total_mcast_qp_attach; 287 int max_ah; 288 int max_fmr; 289 int max_map_per_fmr; 290 int max_srq; 291 int max_srq_wr; 292 int max_srq_sge; 293 unsigned int max_fast_reg_page_list_len; 294 u16 max_pkeys; 295 u8 local_ca_ack_delay; 296 int sig_prot_cap; 297 int sig_guard_cap; 298 struct ib_odp_caps odp_caps; 299 uint64_t timestamp_mask; 300 uint64_t hca_core_clock; /* in KHZ */ 301 }; 302 303 enum ib_mtu { 304 IB_MTU_256 = 1, 305 IB_MTU_512 = 2, 306 IB_MTU_1024 = 3, 307 IB_MTU_2048 = 4, 308 IB_MTU_4096 = 5 309 }; 310 311 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu) 312 { 313 switch (mtu) { 314 case IB_MTU_256: return 256; 315 case IB_MTU_512: return 512; 316 case IB_MTU_1024: return 1024; 317 case IB_MTU_2048: return 2048; 318 case IB_MTU_4096: return 4096; 319 default: return -1; 320 } 321 } 322 323 enum ib_port_state { 324 IB_PORT_NOP = 0, 325 IB_PORT_DOWN = 1, 326 IB_PORT_INIT = 2, 327 IB_PORT_ARMED = 3, 328 IB_PORT_ACTIVE = 4, 329 IB_PORT_ACTIVE_DEFER = 5 330 }; 331 332 enum ib_port_cap_flags { 333 IB_PORT_SM = 1 << 1, 334 IB_PORT_NOTICE_SUP = 1 << 2, 335 IB_PORT_TRAP_SUP = 1 << 3, 336 IB_PORT_OPT_IPD_SUP = 1 << 4, 337 IB_PORT_AUTO_MIGR_SUP = 1 << 5, 338 IB_PORT_SL_MAP_SUP = 1 << 6, 339 IB_PORT_MKEY_NVRAM = 1 << 7, 340 IB_PORT_PKEY_NVRAM = 1 << 8, 341 IB_PORT_LED_INFO_SUP = 1 << 9, 342 IB_PORT_SM_DISABLED = 1 << 10, 343 IB_PORT_SYS_IMAGE_GUID_SUP = 1 << 11, 344 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP = 1 << 12, 345 IB_PORT_EXTENDED_SPEEDS_SUP = 1 << 14, 346 IB_PORT_CM_SUP = 1 << 16, 347 IB_PORT_SNMP_TUNNEL_SUP = 1 << 17, 348 IB_PORT_REINIT_SUP = 1 << 18, 349 IB_PORT_DEVICE_MGMT_SUP = 1 << 19, 350 IB_PORT_VENDOR_CLASS_SUP = 1 << 20, 351 IB_PORT_DR_NOTICE_SUP = 1 << 21, 352 IB_PORT_CAP_MASK_NOTICE_SUP = 1 << 22, 353 IB_PORT_BOOT_MGMT_SUP = 1 << 23, 354 IB_PORT_LINK_LATENCY_SUP = 1 << 24, 355 IB_PORT_CLIENT_REG_SUP = 1 << 25, 356 IB_PORT_IP_BASED_GIDS = 1 << 26, 357 }; 358 359 enum ib_port_width { 360 IB_WIDTH_1X = 1, 361 IB_WIDTH_4X = 2, 362 IB_WIDTH_8X = 4, 363 IB_WIDTH_12X = 8 364 }; 365 366 static inline int ib_width_enum_to_int(enum ib_port_width width) 367 { 368 switch (width) { 369 case IB_WIDTH_1X: return 1; 370 case IB_WIDTH_4X: return 4; 371 case IB_WIDTH_8X: return 8; 372 case IB_WIDTH_12X: return 12; 373 default: return -1; 374 } 375 } 376 377 enum ib_port_speed { 378 IB_SPEED_SDR = 1, 379 IB_SPEED_DDR = 2, 380 IB_SPEED_QDR = 4, 381 IB_SPEED_FDR10 = 8, 382 IB_SPEED_FDR = 16, 383 IB_SPEED_EDR = 32 384 }; 385 386 struct ib_protocol_stats { 387 /* TBD... */ 388 }; 389 390 struct iw_protocol_stats { 391 u64 ipInReceives; 392 u64 ipInHdrErrors; 393 u64 ipInTooBigErrors; 394 u64 ipInNoRoutes; 395 u64 ipInAddrErrors; 396 u64 ipInUnknownProtos; 397 u64 ipInTruncatedPkts; 398 u64 ipInDiscards; 399 u64 ipInDelivers; 400 u64 ipOutForwDatagrams; 401 u64 ipOutRequests; 402 u64 ipOutDiscards; 403 u64 ipOutNoRoutes; 404 u64 ipReasmTimeout; 405 u64 ipReasmReqds; 406 u64 ipReasmOKs; 407 u64 ipReasmFails; 408 u64 ipFragOKs; 409 u64 ipFragFails; 410 u64 ipFragCreates; 411 u64 ipInMcastPkts; 412 u64 ipOutMcastPkts; 413 u64 ipInBcastPkts; 414 u64 ipOutBcastPkts; 415 416 u64 tcpRtoAlgorithm; 417 u64 tcpRtoMin; 418 u64 tcpRtoMax; 419 u64 tcpMaxConn; 420 u64 tcpActiveOpens; 421 u64 tcpPassiveOpens; 422 u64 tcpAttemptFails; 423 u64 tcpEstabResets; 424 u64 tcpCurrEstab; 425 u64 tcpInSegs; 426 u64 tcpOutSegs; 427 u64 tcpRetransSegs; 428 u64 tcpInErrs; 429 u64 tcpOutRsts; 430 }; 431 432 union rdma_protocol_stats { 433 struct ib_protocol_stats ib; 434 struct iw_protocol_stats iw; 435 }; 436 437 /* Define bits for the various functionality this port needs to be supported by 438 * the core. 439 */ 440 /* Management 0x00000FFF */ 441 #define RDMA_CORE_CAP_IB_MAD 0x00000001 442 #define RDMA_CORE_CAP_IB_SMI 0x00000002 443 #define RDMA_CORE_CAP_IB_CM 0x00000004 444 #define RDMA_CORE_CAP_IW_CM 0x00000008 445 #define RDMA_CORE_CAP_IB_SA 0x00000010 446 #define RDMA_CORE_CAP_OPA_MAD 0x00000020 447 448 /* Address format 0x000FF000 */ 449 #define RDMA_CORE_CAP_AF_IB 0x00001000 450 #define RDMA_CORE_CAP_ETH_AH 0x00002000 451 452 /* Protocol 0xFFF00000 */ 453 #define RDMA_CORE_CAP_PROT_IB 0x00100000 454 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000 455 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000 456 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000 457 458 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \ 459 | RDMA_CORE_CAP_IB_MAD \ 460 | RDMA_CORE_CAP_IB_SMI \ 461 | RDMA_CORE_CAP_IB_CM \ 462 | RDMA_CORE_CAP_IB_SA \ 463 | RDMA_CORE_CAP_AF_IB) 464 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \ 465 | RDMA_CORE_CAP_IB_MAD \ 466 | RDMA_CORE_CAP_IB_CM \ 467 | RDMA_CORE_CAP_AF_IB \ 468 | RDMA_CORE_CAP_ETH_AH) 469 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \ 470 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \ 471 | RDMA_CORE_CAP_IB_MAD \ 472 | RDMA_CORE_CAP_IB_CM \ 473 | RDMA_CORE_CAP_AF_IB \ 474 | RDMA_CORE_CAP_ETH_AH) 475 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \ 476 | RDMA_CORE_CAP_IW_CM) 477 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \ 478 | RDMA_CORE_CAP_OPA_MAD) 479 480 struct ib_port_attr { 481 enum ib_port_state state; 482 enum ib_mtu max_mtu; 483 enum ib_mtu active_mtu; 484 int gid_tbl_len; 485 u32 port_cap_flags; 486 u32 max_msg_sz; 487 u32 bad_pkey_cntr; 488 u32 qkey_viol_cntr; 489 u16 pkey_tbl_len; 490 u16 lid; 491 u16 sm_lid; 492 u8 lmc; 493 u8 max_vl_num; 494 u8 sm_sl; 495 u8 subnet_timeout; 496 u8 init_type_reply; 497 u8 active_width; 498 u8 active_speed; 499 u8 phys_state; 500 }; 501 502 enum ib_device_modify_flags { 503 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0, 504 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1 505 }; 506 507 struct ib_device_modify { 508 u64 sys_image_guid; 509 char node_desc[64]; 510 }; 511 512 enum ib_port_modify_flags { 513 IB_PORT_SHUTDOWN = 1, 514 IB_PORT_INIT_TYPE = (1<<2), 515 IB_PORT_RESET_QKEY_CNTR = (1<<3) 516 }; 517 518 struct ib_port_modify { 519 u32 set_port_cap_mask; 520 u32 clr_port_cap_mask; 521 u8 init_type; 522 }; 523 524 enum ib_event_type { 525 IB_EVENT_CQ_ERR, 526 IB_EVENT_QP_FATAL, 527 IB_EVENT_QP_REQ_ERR, 528 IB_EVENT_QP_ACCESS_ERR, 529 IB_EVENT_COMM_EST, 530 IB_EVENT_SQ_DRAINED, 531 IB_EVENT_PATH_MIG, 532 IB_EVENT_PATH_MIG_ERR, 533 IB_EVENT_DEVICE_FATAL, 534 IB_EVENT_PORT_ACTIVE, 535 IB_EVENT_PORT_ERR, 536 IB_EVENT_LID_CHANGE, 537 IB_EVENT_PKEY_CHANGE, 538 IB_EVENT_SM_CHANGE, 539 IB_EVENT_SRQ_ERR, 540 IB_EVENT_SRQ_LIMIT_REACHED, 541 IB_EVENT_QP_LAST_WQE_REACHED, 542 IB_EVENT_CLIENT_REREGISTER, 543 IB_EVENT_GID_CHANGE, 544 }; 545 546 const char *__attribute_const__ ib_event_msg(enum ib_event_type event); 547 548 struct ib_event { 549 struct ib_device *device; 550 union { 551 struct ib_cq *cq; 552 struct ib_qp *qp; 553 struct ib_srq *srq; 554 u8 port_num; 555 } element; 556 enum ib_event_type event; 557 }; 558 559 struct ib_event_handler { 560 struct ib_device *device; 561 void (*handler)(struct ib_event_handler *, struct ib_event *); 562 struct list_head list; 563 }; 564 565 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \ 566 do { \ 567 (_ptr)->device = _device; \ 568 (_ptr)->handler = _handler; \ 569 INIT_LIST_HEAD(&(_ptr)->list); \ 570 } while (0) 571 572 struct ib_global_route { 573 union ib_gid dgid; 574 u32 flow_label; 575 u8 sgid_index; 576 u8 hop_limit; 577 u8 traffic_class; 578 }; 579 580 struct ib_grh { 581 __be32 version_tclass_flow; 582 __be16 paylen; 583 u8 next_hdr; 584 u8 hop_limit; 585 union ib_gid sgid; 586 union ib_gid dgid; 587 }; 588 589 union rdma_network_hdr { 590 struct ib_grh ibgrh; 591 struct { 592 /* The IB spec states that if it's IPv4, the header 593 * is located in the last 20 bytes of the header. 594 */ 595 u8 reserved[20]; 596 struct iphdr roce4grh; 597 }; 598 }; 599 600 enum { 601 IB_MULTICAST_QPN = 0xffffff 602 }; 603 604 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF) 605 606 enum ib_ah_flags { 607 IB_AH_GRH = 1 608 }; 609 610 enum ib_rate { 611 IB_RATE_PORT_CURRENT = 0, 612 IB_RATE_2_5_GBPS = 2, 613 IB_RATE_5_GBPS = 5, 614 IB_RATE_10_GBPS = 3, 615 IB_RATE_20_GBPS = 6, 616 IB_RATE_30_GBPS = 4, 617 IB_RATE_40_GBPS = 7, 618 IB_RATE_60_GBPS = 8, 619 IB_RATE_80_GBPS = 9, 620 IB_RATE_120_GBPS = 10, 621 IB_RATE_14_GBPS = 11, 622 IB_RATE_56_GBPS = 12, 623 IB_RATE_112_GBPS = 13, 624 IB_RATE_168_GBPS = 14, 625 IB_RATE_25_GBPS = 15, 626 IB_RATE_100_GBPS = 16, 627 IB_RATE_200_GBPS = 17, 628 IB_RATE_300_GBPS = 18 629 }; 630 631 /** 632 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the 633 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be 634 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec. 635 * @rate: rate to convert. 636 */ 637 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate); 638 639 /** 640 * ib_rate_to_mbps - Convert the IB rate enum to Mbps. 641 * For example, IB_RATE_2_5_GBPS will be converted to 2500. 642 * @rate: rate to convert. 643 */ 644 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate); 645 646 647 /** 648 * enum ib_mr_type - memory region type 649 * @IB_MR_TYPE_MEM_REG: memory region that is used for 650 * normal registration 651 * @IB_MR_TYPE_SIGNATURE: memory region that is used for 652 * signature operations (data-integrity 653 * capable regions) 654 */ 655 enum ib_mr_type { 656 IB_MR_TYPE_MEM_REG, 657 IB_MR_TYPE_SIGNATURE, 658 }; 659 660 /** 661 * Signature types 662 * IB_SIG_TYPE_NONE: Unprotected. 663 * IB_SIG_TYPE_T10_DIF: Type T10-DIF 664 */ 665 enum ib_signature_type { 666 IB_SIG_TYPE_NONE, 667 IB_SIG_TYPE_T10_DIF, 668 }; 669 670 /** 671 * Signature T10-DIF block-guard types 672 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules. 673 * IB_T10DIF_CSUM: Corresponds to IP checksum rules. 674 */ 675 enum ib_t10_dif_bg_type { 676 IB_T10DIF_CRC, 677 IB_T10DIF_CSUM 678 }; 679 680 /** 681 * struct ib_t10_dif_domain - Parameters specific for T10-DIF 682 * domain. 683 * @bg_type: T10-DIF block guard type (CRC|CSUM) 684 * @pi_interval: protection information interval. 685 * @bg: seed of guard computation. 686 * @app_tag: application tag of guard block 687 * @ref_tag: initial guard block reference tag. 688 * @ref_remap: Indicate wethear the reftag increments each block 689 * @app_escape: Indicate to skip block check if apptag=0xffff 690 * @ref_escape: Indicate to skip block check if reftag=0xffffffff 691 * @apptag_check_mask: check bitmask of application tag. 692 */ 693 struct ib_t10_dif_domain { 694 enum ib_t10_dif_bg_type bg_type; 695 u16 pi_interval; 696 u16 bg; 697 u16 app_tag; 698 u32 ref_tag; 699 bool ref_remap; 700 bool app_escape; 701 bool ref_escape; 702 u16 apptag_check_mask; 703 }; 704 705 /** 706 * struct ib_sig_domain - Parameters for signature domain 707 * @sig_type: specific signauture type 708 * @sig: union of all signature domain attributes that may 709 * be used to set domain layout. 710 */ 711 struct ib_sig_domain { 712 enum ib_signature_type sig_type; 713 union { 714 struct ib_t10_dif_domain dif; 715 } sig; 716 }; 717 718 /** 719 * struct ib_sig_attrs - Parameters for signature handover operation 720 * @check_mask: bitmask for signature byte check (8 bytes) 721 * @mem: memory domain layout desciptor. 722 * @wire: wire domain layout desciptor. 723 */ 724 struct ib_sig_attrs { 725 u8 check_mask; 726 struct ib_sig_domain mem; 727 struct ib_sig_domain wire; 728 }; 729 730 enum ib_sig_err_type { 731 IB_SIG_BAD_GUARD, 732 IB_SIG_BAD_REFTAG, 733 IB_SIG_BAD_APPTAG, 734 }; 735 736 /** 737 * struct ib_sig_err - signature error descriptor 738 */ 739 struct ib_sig_err { 740 enum ib_sig_err_type err_type; 741 u32 expected; 742 u32 actual; 743 u64 sig_err_offset; 744 u32 key; 745 }; 746 747 enum ib_mr_status_check { 748 IB_MR_CHECK_SIG_STATUS = 1, 749 }; 750 751 /** 752 * struct ib_mr_status - Memory region status container 753 * 754 * @fail_status: Bitmask of MR checks status. For each 755 * failed check a corresponding status bit is set. 756 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS 757 * failure. 758 */ 759 struct ib_mr_status { 760 u32 fail_status; 761 struct ib_sig_err sig_err; 762 }; 763 764 /** 765 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate 766 * enum. 767 * @mult: multiple to convert. 768 */ 769 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult); 770 771 struct ib_ah_attr { 772 struct ib_global_route grh; 773 u16 dlid; 774 u8 sl; 775 u8 src_path_bits; 776 u8 static_rate; 777 u8 ah_flags; 778 u8 port_num; 779 u8 dmac[ETH_ALEN]; 780 }; 781 782 enum ib_wc_status { 783 IB_WC_SUCCESS, 784 IB_WC_LOC_LEN_ERR, 785 IB_WC_LOC_QP_OP_ERR, 786 IB_WC_LOC_EEC_OP_ERR, 787 IB_WC_LOC_PROT_ERR, 788 IB_WC_WR_FLUSH_ERR, 789 IB_WC_MW_BIND_ERR, 790 IB_WC_BAD_RESP_ERR, 791 IB_WC_LOC_ACCESS_ERR, 792 IB_WC_REM_INV_REQ_ERR, 793 IB_WC_REM_ACCESS_ERR, 794 IB_WC_REM_OP_ERR, 795 IB_WC_RETRY_EXC_ERR, 796 IB_WC_RNR_RETRY_EXC_ERR, 797 IB_WC_LOC_RDD_VIOL_ERR, 798 IB_WC_REM_INV_RD_REQ_ERR, 799 IB_WC_REM_ABORT_ERR, 800 IB_WC_INV_EECN_ERR, 801 IB_WC_INV_EEC_STATE_ERR, 802 IB_WC_FATAL_ERR, 803 IB_WC_RESP_TIMEOUT_ERR, 804 IB_WC_GENERAL_ERR 805 }; 806 807 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status); 808 809 enum ib_wc_opcode { 810 IB_WC_SEND, 811 IB_WC_RDMA_WRITE, 812 IB_WC_RDMA_READ, 813 IB_WC_COMP_SWAP, 814 IB_WC_FETCH_ADD, 815 IB_WC_LSO, 816 IB_WC_LOCAL_INV, 817 IB_WC_REG_MR, 818 IB_WC_MASKED_COMP_SWAP, 819 IB_WC_MASKED_FETCH_ADD, 820 /* 821 * Set value of IB_WC_RECV so consumers can test if a completion is a 822 * receive by testing (opcode & IB_WC_RECV). 823 */ 824 IB_WC_RECV = 1 << 7, 825 IB_WC_RECV_RDMA_WITH_IMM 826 }; 827 828 enum ib_wc_flags { 829 IB_WC_GRH = 1, 830 IB_WC_WITH_IMM = (1<<1), 831 IB_WC_WITH_INVALIDATE = (1<<2), 832 IB_WC_IP_CSUM_OK = (1<<3), 833 IB_WC_WITH_SMAC = (1<<4), 834 IB_WC_WITH_VLAN = (1<<5), 835 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6), 836 }; 837 838 struct ib_wc { 839 union { 840 u64 wr_id; 841 struct ib_cqe *wr_cqe; 842 }; 843 enum ib_wc_status status; 844 enum ib_wc_opcode opcode; 845 u32 vendor_err; 846 u32 byte_len; 847 struct ib_qp *qp; 848 union { 849 __be32 imm_data; 850 u32 invalidate_rkey; 851 } ex; 852 u32 src_qp; 853 int wc_flags; 854 u16 pkey_index; 855 u16 slid; 856 u8 sl; 857 u8 dlid_path_bits; 858 u8 port_num; /* valid only for DR SMPs on switches */ 859 u8 smac[ETH_ALEN]; 860 u16 vlan_id; 861 u8 network_hdr_type; 862 }; 863 864 enum ib_cq_notify_flags { 865 IB_CQ_SOLICITED = 1 << 0, 866 IB_CQ_NEXT_COMP = 1 << 1, 867 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP, 868 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2, 869 }; 870 871 enum ib_srq_type { 872 IB_SRQT_BASIC, 873 IB_SRQT_XRC 874 }; 875 876 enum ib_srq_attr_mask { 877 IB_SRQ_MAX_WR = 1 << 0, 878 IB_SRQ_LIMIT = 1 << 1, 879 }; 880 881 struct ib_srq_attr { 882 u32 max_wr; 883 u32 max_sge; 884 u32 srq_limit; 885 }; 886 887 struct ib_srq_init_attr { 888 void (*event_handler)(struct ib_event *, void *); 889 void *srq_context; 890 struct ib_srq_attr attr; 891 enum ib_srq_type srq_type; 892 893 union { 894 struct { 895 struct ib_xrcd *xrcd; 896 struct ib_cq *cq; 897 } xrc; 898 } ext; 899 }; 900 901 struct ib_qp_cap { 902 u32 max_send_wr; 903 u32 max_recv_wr; 904 u32 max_send_sge; 905 u32 max_recv_sge; 906 u32 max_inline_data; 907 }; 908 909 enum ib_sig_type { 910 IB_SIGNAL_ALL_WR, 911 IB_SIGNAL_REQ_WR 912 }; 913 914 enum ib_qp_type { 915 /* 916 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries 917 * here (and in that order) since the MAD layer uses them as 918 * indices into a 2-entry table. 919 */ 920 IB_QPT_SMI, 921 IB_QPT_GSI, 922 923 IB_QPT_RC, 924 IB_QPT_UC, 925 IB_QPT_UD, 926 IB_QPT_RAW_IPV6, 927 IB_QPT_RAW_ETHERTYPE, 928 IB_QPT_RAW_PACKET = 8, 929 IB_QPT_XRC_INI = 9, 930 IB_QPT_XRC_TGT, 931 IB_QPT_MAX, 932 /* Reserve a range for qp types internal to the low level driver. 933 * These qp types will not be visible at the IB core layer, so the 934 * IB_QPT_MAX usages should not be affected in the core layer 935 */ 936 IB_QPT_RESERVED1 = 0x1000, 937 IB_QPT_RESERVED2, 938 IB_QPT_RESERVED3, 939 IB_QPT_RESERVED4, 940 IB_QPT_RESERVED5, 941 IB_QPT_RESERVED6, 942 IB_QPT_RESERVED7, 943 IB_QPT_RESERVED8, 944 IB_QPT_RESERVED9, 945 IB_QPT_RESERVED10, 946 }; 947 948 enum ib_qp_create_flags { 949 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0, 950 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1, 951 IB_QP_CREATE_NETIF_QP = 1 << 5, 952 IB_QP_CREATE_SIGNATURE_EN = 1 << 6, 953 IB_QP_CREATE_USE_GFP_NOIO = 1 << 7, 954 /* reserve bits 26-31 for low level drivers' internal use */ 955 IB_QP_CREATE_RESERVED_START = 1 << 26, 956 IB_QP_CREATE_RESERVED_END = 1 << 31, 957 }; 958 959 /* 960 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler 961 * callback to destroy the passed in QP. 962 */ 963 964 struct ib_qp_init_attr { 965 void (*event_handler)(struct ib_event *, void *); 966 void *qp_context; 967 struct ib_cq *send_cq; 968 struct ib_cq *recv_cq; 969 struct ib_srq *srq; 970 struct ib_xrcd *xrcd; /* XRC TGT QPs only */ 971 struct ib_qp_cap cap; 972 enum ib_sig_type sq_sig_type; 973 enum ib_qp_type qp_type; 974 enum ib_qp_create_flags create_flags; 975 u8 port_num; /* special QP types only */ 976 }; 977 978 struct ib_qp_open_attr { 979 void (*event_handler)(struct ib_event *, void *); 980 void *qp_context; 981 u32 qp_num; 982 enum ib_qp_type qp_type; 983 }; 984 985 enum ib_rnr_timeout { 986 IB_RNR_TIMER_655_36 = 0, 987 IB_RNR_TIMER_000_01 = 1, 988 IB_RNR_TIMER_000_02 = 2, 989 IB_RNR_TIMER_000_03 = 3, 990 IB_RNR_TIMER_000_04 = 4, 991 IB_RNR_TIMER_000_06 = 5, 992 IB_RNR_TIMER_000_08 = 6, 993 IB_RNR_TIMER_000_12 = 7, 994 IB_RNR_TIMER_000_16 = 8, 995 IB_RNR_TIMER_000_24 = 9, 996 IB_RNR_TIMER_000_32 = 10, 997 IB_RNR_TIMER_000_48 = 11, 998 IB_RNR_TIMER_000_64 = 12, 999 IB_RNR_TIMER_000_96 = 13, 1000 IB_RNR_TIMER_001_28 = 14, 1001 IB_RNR_TIMER_001_92 = 15, 1002 IB_RNR_TIMER_002_56 = 16, 1003 IB_RNR_TIMER_003_84 = 17, 1004 IB_RNR_TIMER_005_12 = 18, 1005 IB_RNR_TIMER_007_68 = 19, 1006 IB_RNR_TIMER_010_24 = 20, 1007 IB_RNR_TIMER_015_36 = 21, 1008 IB_RNR_TIMER_020_48 = 22, 1009 IB_RNR_TIMER_030_72 = 23, 1010 IB_RNR_TIMER_040_96 = 24, 1011 IB_RNR_TIMER_061_44 = 25, 1012 IB_RNR_TIMER_081_92 = 26, 1013 IB_RNR_TIMER_122_88 = 27, 1014 IB_RNR_TIMER_163_84 = 28, 1015 IB_RNR_TIMER_245_76 = 29, 1016 IB_RNR_TIMER_327_68 = 30, 1017 IB_RNR_TIMER_491_52 = 31 1018 }; 1019 1020 enum ib_qp_attr_mask { 1021 IB_QP_STATE = 1, 1022 IB_QP_CUR_STATE = (1<<1), 1023 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2), 1024 IB_QP_ACCESS_FLAGS = (1<<3), 1025 IB_QP_PKEY_INDEX = (1<<4), 1026 IB_QP_PORT = (1<<5), 1027 IB_QP_QKEY = (1<<6), 1028 IB_QP_AV = (1<<7), 1029 IB_QP_PATH_MTU = (1<<8), 1030 IB_QP_TIMEOUT = (1<<9), 1031 IB_QP_RETRY_CNT = (1<<10), 1032 IB_QP_RNR_RETRY = (1<<11), 1033 IB_QP_RQ_PSN = (1<<12), 1034 IB_QP_MAX_QP_RD_ATOMIC = (1<<13), 1035 IB_QP_ALT_PATH = (1<<14), 1036 IB_QP_MIN_RNR_TIMER = (1<<15), 1037 IB_QP_SQ_PSN = (1<<16), 1038 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17), 1039 IB_QP_PATH_MIG_STATE = (1<<18), 1040 IB_QP_CAP = (1<<19), 1041 IB_QP_DEST_QPN = (1<<20), 1042 IB_QP_RESERVED1 = (1<<21), 1043 IB_QP_RESERVED2 = (1<<22), 1044 IB_QP_RESERVED3 = (1<<23), 1045 IB_QP_RESERVED4 = (1<<24), 1046 }; 1047 1048 enum ib_qp_state { 1049 IB_QPS_RESET, 1050 IB_QPS_INIT, 1051 IB_QPS_RTR, 1052 IB_QPS_RTS, 1053 IB_QPS_SQD, 1054 IB_QPS_SQE, 1055 IB_QPS_ERR 1056 }; 1057 1058 enum ib_mig_state { 1059 IB_MIG_MIGRATED, 1060 IB_MIG_REARM, 1061 IB_MIG_ARMED 1062 }; 1063 1064 enum ib_mw_type { 1065 IB_MW_TYPE_1 = 1, 1066 IB_MW_TYPE_2 = 2 1067 }; 1068 1069 struct ib_qp_attr { 1070 enum ib_qp_state qp_state; 1071 enum ib_qp_state cur_qp_state; 1072 enum ib_mtu path_mtu; 1073 enum ib_mig_state path_mig_state; 1074 u32 qkey; 1075 u32 rq_psn; 1076 u32 sq_psn; 1077 u32 dest_qp_num; 1078 int qp_access_flags; 1079 struct ib_qp_cap cap; 1080 struct ib_ah_attr ah_attr; 1081 struct ib_ah_attr alt_ah_attr; 1082 u16 pkey_index; 1083 u16 alt_pkey_index; 1084 u8 en_sqd_async_notify; 1085 u8 sq_draining; 1086 u8 max_rd_atomic; 1087 u8 max_dest_rd_atomic; 1088 u8 min_rnr_timer; 1089 u8 port_num; 1090 u8 timeout; 1091 u8 retry_cnt; 1092 u8 rnr_retry; 1093 u8 alt_port_num; 1094 u8 alt_timeout; 1095 }; 1096 1097 enum ib_wr_opcode { 1098 IB_WR_RDMA_WRITE, 1099 IB_WR_RDMA_WRITE_WITH_IMM, 1100 IB_WR_SEND, 1101 IB_WR_SEND_WITH_IMM, 1102 IB_WR_RDMA_READ, 1103 IB_WR_ATOMIC_CMP_AND_SWP, 1104 IB_WR_ATOMIC_FETCH_AND_ADD, 1105 IB_WR_LSO, 1106 IB_WR_SEND_WITH_INV, 1107 IB_WR_RDMA_READ_WITH_INV, 1108 IB_WR_LOCAL_INV, 1109 IB_WR_REG_MR, 1110 IB_WR_MASKED_ATOMIC_CMP_AND_SWP, 1111 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD, 1112 IB_WR_REG_SIG_MR, 1113 /* reserve values for low level drivers' internal use. 1114 * These values will not be used at all in the ib core layer. 1115 */ 1116 IB_WR_RESERVED1 = 0xf0, 1117 IB_WR_RESERVED2, 1118 IB_WR_RESERVED3, 1119 IB_WR_RESERVED4, 1120 IB_WR_RESERVED5, 1121 IB_WR_RESERVED6, 1122 IB_WR_RESERVED7, 1123 IB_WR_RESERVED8, 1124 IB_WR_RESERVED9, 1125 IB_WR_RESERVED10, 1126 }; 1127 1128 enum ib_send_flags { 1129 IB_SEND_FENCE = 1, 1130 IB_SEND_SIGNALED = (1<<1), 1131 IB_SEND_SOLICITED = (1<<2), 1132 IB_SEND_INLINE = (1<<3), 1133 IB_SEND_IP_CSUM = (1<<4), 1134 1135 /* reserve bits 26-31 for low level drivers' internal use */ 1136 IB_SEND_RESERVED_START = (1 << 26), 1137 IB_SEND_RESERVED_END = (1 << 31), 1138 }; 1139 1140 struct ib_sge { 1141 u64 addr; 1142 u32 length; 1143 u32 lkey; 1144 }; 1145 1146 struct ib_cqe { 1147 void (*done)(struct ib_cq *cq, struct ib_wc *wc); 1148 }; 1149 1150 struct ib_send_wr { 1151 struct ib_send_wr *next; 1152 union { 1153 u64 wr_id; 1154 struct ib_cqe *wr_cqe; 1155 }; 1156 struct ib_sge *sg_list; 1157 int num_sge; 1158 enum ib_wr_opcode opcode; 1159 int send_flags; 1160 union { 1161 __be32 imm_data; 1162 u32 invalidate_rkey; 1163 } ex; 1164 }; 1165 1166 struct ib_rdma_wr { 1167 struct ib_send_wr wr; 1168 u64 remote_addr; 1169 u32 rkey; 1170 }; 1171 1172 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr) 1173 { 1174 return container_of(wr, struct ib_rdma_wr, wr); 1175 } 1176 1177 struct ib_atomic_wr { 1178 struct ib_send_wr wr; 1179 u64 remote_addr; 1180 u64 compare_add; 1181 u64 swap; 1182 u64 compare_add_mask; 1183 u64 swap_mask; 1184 u32 rkey; 1185 }; 1186 1187 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr) 1188 { 1189 return container_of(wr, struct ib_atomic_wr, wr); 1190 } 1191 1192 struct ib_ud_wr { 1193 struct ib_send_wr wr; 1194 struct ib_ah *ah; 1195 void *header; 1196 int hlen; 1197 int mss; 1198 u32 remote_qpn; 1199 u32 remote_qkey; 1200 u16 pkey_index; /* valid for GSI only */ 1201 u8 port_num; /* valid for DR SMPs on switch only */ 1202 }; 1203 1204 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr) 1205 { 1206 return container_of(wr, struct ib_ud_wr, wr); 1207 } 1208 1209 struct ib_reg_wr { 1210 struct ib_send_wr wr; 1211 struct ib_mr *mr; 1212 u32 key; 1213 int access; 1214 }; 1215 1216 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr) 1217 { 1218 return container_of(wr, struct ib_reg_wr, wr); 1219 } 1220 1221 struct ib_sig_handover_wr { 1222 struct ib_send_wr wr; 1223 struct ib_sig_attrs *sig_attrs; 1224 struct ib_mr *sig_mr; 1225 int access_flags; 1226 struct ib_sge *prot; 1227 }; 1228 1229 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr) 1230 { 1231 return container_of(wr, struct ib_sig_handover_wr, wr); 1232 } 1233 1234 struct ib_recv_wr { 1235 struct ib_recv_wr *next; 1236 union { 1237 u64 wr_id; 1238 struct ib_cqe *wr_cqe; 1239 }; 1240 struct ib_sge *sg_list; 1241 int num_sge; 1242 }; 1243 1244 enum ib_access_flags { 1245 IB_ACCESS_LOCAL_WRITE = 1, 1246 IB_ACCESS_REMOTE_WRITE = (1<<1), 1247 IB_ACCESS_REMOTE_READ = (1<<2), 1248 IB_ACCESS_REMOTE_ATOMIC = (1<<3), 1249 IB_ACCESS_MW_BIND = (1<<4), 1250 IB_ZERO_BASED = (1<<5), 1251 IB_ACCESS_ON_DEMAND = (1<<6), 1252 }; 1253 1254 struct ib_phys_buf { 1255 u64 addr; 1256 u64 size; 1257 }; 1258 1259 /* 1260 * XXX: these are apparently used for ->rereg_user_mr, no idea why they 1261 * are hidden here instead of a uapi header! 1262 */ 1263 enum ib_mr_rereg_flags { 1264 IB_MR_REREG_TRANS = 1, 1265 IB_MR_REREG_PD = (1<<1), 1266 IB_MR_REREG_ACCESS = (1<<2), 1267 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1) 1268 }; 1269 1270 struct ib_fmr_attr { 1271 int max_pages; 1272 int max_maps; 1273 u8 page_shift; 1274 }; 1275 1276 struct ib_umem; 1277 1278 struct ib_ucontext { 1279 struct ib_device *device; 1280 struct list_head pd_list; 1281 struct list_head mr_list; 1282 struct list_head mw_list; 1283 struct list_head cq_list; 1284 struct list_head qp_list; 1285 struct list_head srq_list; 1286 struct list_head ah_list; 1287 struct list_head xrcd_list; 1288 struct list_head rule_list; 1289 int closing; 1290 1291 struct pid *tgid; 1292 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING 1293 struct rb_root umem_tree; 1294 /* 1295 * Protects .umem_rbroot and tree, as well as odp_mrs_count and 1296 * mmu notifiers registration. 1297 */ 1298 struct rw_semaphore umem_rwsem; 1299 void (*invalidate_range)(struct ib_umem *umem, 1300 unsigned long start, unsigned long end); 1301 1302 struct mmu_notifier mn; 1303 atomic_t notifier_count; 1304 /* A list of umems that don't have private mmu notifier counters yet. */ 1305 struct list_head no_private_counters; 1306 int odp_mrs_count; 1307 #endif 1308 }; 1309 1310 struct ib_uobject { 1311 u64 user_handle; /* handle given to us by userspace */ 1312 struct ib_ucontext *context; /* associated user context */ 1313 void *object; /* containing object */ 1314 struct list_head list; /* link to context's list */ 1315 int id; /* index into kernel idr */ 1316 struct kref ref; 1317 struct rw_semaphore mutex; /* protects .live */ 1318 struct rcu_head rcu; /* kfree_rcu() overhead */ 1319 int live; 1320 }; 1321 1322 struct ib_udata { 1323 const void __user *inbuf; 1324 void __user *outbuf; 1325 size_t inlen; 1326 size_t outlen; 1327 }; 1328 1329 struct ib_pd { 1330 u32 local_dma_lkey; 1331 struct ib_device *device; 1332 struct ib_uobject *uobject; 1333 atomic_t usecnt; /* count all resources */ 1334 struct ib_mr *local_mr; 1335 }; 1336 1337 struct ib_xrcd { 1338 struct ib_device *device; 1339 atomic_t usecnt; /* count all exposed resources */ 1340 struct inode *inode; 1341 1342 struct mutex tgt_qp_mutex; 1343 struct list_head tgt_qp_list; 1344 }; 1345 1346 struct ib_ah { 1347 struct ib_device *device; 1348 struct ib_pd *pd; 1349 struct ib_uobject *uobject; 1350 }; 1351 1352 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context); 1353 1354 enum ib_poll_context { 1355 IB_POLL_DIRECT, /* caller context, no hw completions */ 1356 IB_POLL_SOFTIRQ, /* poll from softirq context */ 1357 IB_POLL_WORKQUEUE, /* poll from workqueue */ 1358 }; 1359 1360 struct ib_cq { 1361 struct ib_device *device; 1362 struct ib_uobject *uobject; 1363 ib_comp_handler comp_handler; 1364 void (*event_handler)(struct ib_event *, void *); 1365 void *cq_context; 1366 int cqe; 1367 atomic_t usecnt; /* count number of work queues */ 1368 enum ib_poll_context poll_ctx; 1369 struct ib_wc *wc; 1370 union { 1371 struct irq_poll iop; 1372 struct work_struct work; 1373 }; 1374 }; 1375 1376 struct ib_srq { 1377 struct ib_device *device; 1378 struct ib_pd *pd; 1379 struct ib_uobject *uobject; 1380 void (*event_handler)(struct ib_event *, void *); 1381 void *srq_context; 1382 enum ib_srq_type srq_type; 1383 atomic_t usecnt; 1384 1385 union { 1386 struct { 1387 struct ib_xrcd *xrcd; 1388 struct ib_cq *cq; 1389 u32 srq_num; 1390 } xrc; 1391 } ext; 1392 }; 1393 1394 struct ib_qp { 1395 struct ib_device *device; 1396 struct ib_pd *pd; 1397 struct ib_cq *send_cq; 1398 struct ib_cq *recv_cq; 1399 struct ib_srq *srq; 1400 struct ib_xrcd *xrcd; /* XRC TGT QPs only */ 1401 struct list_head xrcd_list; 1402 /* count times opened, mcast attaches, flow attaches */ 1403 atomic_t usecnt; 1404 struct list_head open_list; 1405 struct ib_qp *real_qp; 1406 struct ib_uobject *uobject; 1407 void (*event_handler)(struct ib_event *, void *); 1408 void *qp_context; 1409 u32 qp_num; 1410 enum ib_qp_type qp_type; 1411 }; 1412 1413 struct ib_mr { 1414 struct ib_device *device; 1415 struct ib_pd *pd; 1416 struct ib_uobject *uobject; 1417 u32 lkey; 1418 u32 rkey; 1419 u64 iova; 1420 u32 length; 1421 unsigned int page_size; 1422 atomic_t usecnt; /* count number of MWs */ 1423 }; 1424 1425 struct ib_mw { 1426 struct ib_device *device; 1427 struct ib_pd *pd; 1428 struct ib_uobject *uobject; 1429 u32 rkey; 1430 enum ib_mw_type type; 1431 }; 1432 1433 struct ib_fmr { 1434 struct ib_device *device; 1435 struct ib_pd *pd; 1436 struct list_head list; 1437 u32 lkey; 1438 u32 rkey; 1439 }; 1440 1441 /* Supported steering options */ 1442 enum ib_flow_attr_type { 1443 /* steering according to rule specifications */ 1444 IB_FLOW_ATTR_NORMAL = 0x0, 1445 /* default unicast and multicast rule - 1446 * receive all Eth traffic which isn't steered to any QP 1447 */ 1448 IB_FLOW_ATTR_ALL_DEFAULT = 0x1, 1449 /* default multicast rule - 1450 * receive all Eth multicast traffic which isn't steered to any QP 1451 */ 1452 IB_FLOW_ATTR_MC_DEFAULT = 0x2, 1453 /* sniffer rule - receive all port traffic */ 1454 IB_FLOW_ATTR_SNIFFER = 0x3 1455 }; 1456 1457 /* Supported steering header types */ 1458 enum ib_flow_spec_type { 1459 /* L2 headers*/ 1460 IB_FLOW_SPEC_ETH = 0x20, 1461 IB_FLOW_SPEC_IB = 0x22, 1462 /* L3 header*/ 1463 IB_FLOW_SPEC_IPV4 = 0x30, 1464 /* L4 headers*/ 1465 IB_FLOW_SPEC_TCP = 0x40, 1466 IB_FLOW_SPEC_UDP = 0x41 1467 }; 1468 #define IB_FLOW_SPEC_LAYER_MASK 0xF0 1469 #define IB_FLOW_SPEC_SUPPORT_LAYERS 4 1470 1471 /* Flow steering rule priority is set according to it's domain. 1472 * Lower domain value means higher priority. 1473 */ 1474 enum ib_flow_domain { 1475 IB_FLOW_DOMAIN_USER, 1476 IB_FLOW_DOMAIN_ETHTOOL, 1477 IB_FLOW_DOMAIN_RFS, 1478 IB_FLOW_DOMAIN_NIC, 1479 IB_FLOW_DOMAIN_NUM /* Must be last */ 1480 }; 1481 1482 struct ib_flow_eth_filter { 1483 u8 dst_mac[6]; 1484 u8 src_mac[6]; 1485 __be16 ether_type; 1486 __be16 vlan_tag; 1487 }; 1488 1489 struct ib_flow_spec_eth { 1490 enum ib_flow_spec_type type; 1491 u16 size; 1492 struct ib_flow_eth_filter val; 1493 struct ib_flow_eth_filter mask; 1494 }; 1495 1496 struct ib_flow_ib_filter { 1497 __be16 dlid; 1498 __u8 sl; 1499 }; 1500 1501 struct ib_flow_spec_ib { 1502 enum ib_flow_spec_type type; 1503 u16 size; 1504 struct ib_flow_ib_filter val; 1505 struct ib_flow_ib_filter mask; 1506 }; 1507 1508 struct ib_flow_ipv4_filter { 1509 __be32 src_ip; 1510 __be32 dst_ip; 1511 }; 1512 1513 struct ib_flow_spec_ipv4 { 1514 enum ib_flow_spec_type type; 1515 u16 size; 1516 struct ib_flow_ipv4_filter val; 1517 struct ib_flow_ipv4_filter mask; 1518 }; 1519 1520 struct ib_flow_tcp_udp_filter { 1521 __be16 dst_port; 1522 __be16 src_port; 1523 }; 1524 1525 struct ib_flow_spec_tcp_udp { 1526 enum ib_flow_spec_type type; 1527 u16 size; 1528 struct ib_flow_tcp_udp_filter val; 1529 struct ib_flow_tcp_udp_filter mask; 1530 }; 1531 1532 union ib_flow_spec { 1533 struct { 1534 enum ib_flow_spec_type type; 1535 u16 size; 1536 }; 1537 struct ib_flow_spec_eth eth; 1538 struct ib_flow_spec_ib ib; 1539 struct ib_flow_spec_ipv4 ipv4; 1540 struct ib_flow_spec_tcp_udp tcp_udp; 1541 }; 1542 1543 struct ib_flow_attr { 1544 enum ib_flow_attr_type type; 1545 u16 size; 1546 u16 priority; 1547 u32 flags; 1548 u8 num_of_specs; 1549 u8 port; 1550 /* Following are the optional layers according to user request 1551 * struct ib_flow_spec_xxx 1552 * struct ib_flow_spec_yyy 1553 */ 1554 }; 1555 1556 struct ib_flow { 1557 struct ib_qp *qp; 1558 struct ib_uobject *uobject; 1559 }; 1560 1561 struct ib_mad_hdr; 1562 struct ib_grh; 1563 1564 enum ib_process_mad_flags { 1565 IB_MAD_IGNORE_MKEY = 1, 1566 IB_MAD_IGNORE_BKEY = 2, 1567 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY 1568 }; 1569 1570 enum ib_mad_result { 1571 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */ 1572 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */ 1573 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */ 1574 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */ 1575 }; 1576 1577 #define IB_DEVICE_NAME_MAX 64 1578 1579 struct ib_cache { 1580 rwlock_t lock; 1581 struct ib_event_handler event_handler; 1582 struct ib_pkey_cache **pkey_cache; 1583 struct ib_gid_table **gid_cache; 1584 u8 *lmc_cache; 1585 }; 1586 1587 struct ib_dma_mapping_ops { 1588 int (*mapping_error)(struct ib_device *dev, 1589 u64 dma_addr); 1590 u64 (*map_single)(struct ib_device *dev, 1591 void *ptr, size_t size, 1592 enum dma_data_direction direction); 1593 void (*unmap_single)(struct ib_device *dev, 1594 u64 addr, size_t size, 1595 enum dma_data_direction direction); 1596 u64 (*map_page)(struct ib_device *dev, 1597 struct page *page, unsigned long offset, 1598 size_t size, 1599 enum dma_data_direction direction); 1600 void (*unmap_page)(struct ib_device *dev, 1601 u64 addr, size_t size, 1602 enum dma_data_direction direction); 1603 int (*map_sg)(struct ib_device *dev, 1604 struct scatterlist *sg, int nents, 1605 enum dma_data_direction direction); 1606 void (*unmap_sg)(struct ib_device *dev, 1607 struct scatterlist *sg, int nents, 1608 enum dma_data_direction direction); 1609 void (*sync_single_for_cpu)(struct ib_device *dev, 1610 u64 dma_handle, 1611 size_t size, 1612 enum dma_data_direction dir); 1613 void (*sync_single_for_device)(struct ib_device *dev, 1614 u64 dma_handle, 1615 size_t size, 1616 enum dma_data_direction dir); 1617 void *(*alloc_coherent)(struct ib_device *dev, 1618 size_t size, 1619 u64 *dma_handle, 1620 gfp_t flag); 1621 void (*free_coherent)(struct ib_device *dev, 1622 size_t size, void *cpu_addr, 1623 u64 dma_handle); 1624 }; 1625 1626 struct iw_cm_verbs; 1627 1628 struct ib_port_immutable { 1629 int pkey_tbl_len; 1630 int gid_tbl_len; 1631 u32 core_cap_flags; 1632 u32 max_mad_size; 1633 }; 1634 1635 struct ib_device { 1636 struct device *dma_device; 1637 1638 char name[IB_DEVICE_NAME_MAX]; 1639 1640 struct list_head event_handler_list; 1641 spinlock_t event_handler_lock; 1642 1643 spinlock_t client_data_lock; 1644 struct list_head core_list; 1645 /* Access to the client_data_list is protected by the client_data_lock 1646 * spinlock and the lists_rwsem read-write semaphore */ 1647 struct list_head client_data_list; 1648 1649 struct ib_cache cache; 1650 /** 1651 * port_immutable is indexed by port number 1652 */ 1653 struct ib_port_immutable *port_immutable; 1654 1655 int num_comp_vectors; 1656 1657 struct iw_cm_verbs *iwcm; 1658 1659 int (*get_protocol_stats)(struct ib_device *device, 1660 union rdma_protocol_stats *stats); 1661 int (*query_device)(struct ib_device *device, 1662 struct ib_device_attr *device_attr, 1663 struct ib_udata *udata); 1664 int (*query_port)(struct ib_device *device, 1665 u8 port_num, 1666 struct ib_port_attr *port_attr); 1667 enum rdma_link_layer (*get_link_layer)(struct ib_device *device, 1668 u8 port_num); 1669 /* When calling get_netdev, the HW vendor's driver should return the 1670 * net device of device @device at port @port_num or NULL if such 1671 * a net device doesn't exist. The vendor driver should call dev_hold 1672 * on this net device. The HW vendor's device driver must guarantee 1673 * that this function returns NULL before the net device reaches 1674 * NETDEV_UNREGISTER_FINAL state. 1675 */ 1676 struct net_device *(*get_netdev)(struct ib_device *device, 1677 u8 port_num); 1678 int (*query_gid)(struct ib_device *device, 1679 u8 port_num, int index, 1680 union ib_gid *gid); 1681 /* When calling add_gid, the HW vendor's driver should 1682 * add the gid of device @device at gid index @index of 1683 * port @port_num to be @gid. Meta-info of that gid (for example, 1684 * the network device related to this gid is available 1685 * at @attr. @context allows the HW vendor driver to store extra 1686 * information together with a GID entry. The HW vendor may allocate 1687 * memory to contain this information and store it in @context when a 1688 * new GID entry is written to. Params are consistent until the next 1689 * call of add_gid or delete_gid. The function should return 0 on 1690 * success or error otherwise. The function could be called 1691 * concurrently for different ports. This function is only called 1692 * when roce_gid_table is used. 1693 */ 1694 int (*add_gid)(struct ib_device *device, 1695 u8 port_num, 1696 unsigned int index, 1697 const union ib_gid *gid, 1698 const struct ib_gid_attr *attr, 1699 void **context); 1700 /* When calling del_gid, the HW vendor's driver should delete the 1701 * gid of device @device at gid index @index of port @port_num. 1702 * Upon the deletion of a GID entry, the HW vendor must free any 1703 * allocated memory. The caller will clear @context afterwards. 1704 * This function is only called when roce_gid_table is used. 1705 */ 1706 int (*del_gid)(struct ib_device *device, 1707 u8 port_num, 1708 unsigned int index, 1709 void **context); 1710 int (*query_pkey)(struct ib_device *device, 1711 u8 port_num, u16 index, u16 *pkey); 1712 int (*modify_device)(struct ib_device *device, 1713 int device_modify_mask, 1714 struct ib_device_modify *device_modify); 1715 int (*modify_port)(struct ib_device *device, 1716 u8 port_num, int port_modify_mask, 1717 struct ib_port_modify *port_modify); 1718 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device, 1719 struct ib_udata *udata); 1720 int (*dealloc_ucontext)(struct ib_ucontext *context); 1721 int (*mmap)(struct ib_ucontext *context, 1722 struct vm_area_struct *vma); 1723 struct ib_pd * (*alloc_pd)(struct ib_device *device, 1724 struct ib_ucontext *context, 1725 struct ib_udata *udata); 1726 int (*dealloc_pd)(struct ib_pd *pd); 1727 struct ib_ah * (*create_ah)(struct ib_pd *pd, 1728 struct ib_ah_attr *ah_attr); 1729 int (*modify_ah)(struct ib_ah *ah, 1730 struct ib_ah_attr *ah_attr); 1731 int (*query_ah)(struct ib_ah *ah, 1732 struct ib_ah_attr *ah_attr); 1733 int (*destroy_ah)(struct ib_ah *ah); 1734 struct ib_srq * (*create_srq)(struct ib_pd *pd, 1735 struct ib_srq_init_attr *srq_init_attr, 1736 struct ib_udata *udata); 1737 int (*modify_srq)(struct ib_srq *srq, 1738 struct ib_srq_attr *srq_attr, 1739 enum ib_srq_attr_mask srq_attr_mask, 1740 struct ib_udata *udata); 1741 int (*query_srq)(struct ib_srq *srq, 1742 struct ib_srq_attr *srq_attr); 1743 int (*destroy_srq)(struct ib_srq *srq); 1744 int (*post_srq_recv)(struct ib_srq *srq, 1745 struct ib_recv_wr *recv_wr, 1746 struct ib_recv_wr **bad_recv_wr); 1747 struct ib_qp * (*create_qp)(struct ib_pd *pd, 1748 struct ib_qp_init_attr *qp_init_attr, 1749 struct ib_udata *udata); 1750 int (*modify_qp)(struct ib_qp *qp, 1751 struct ib_qp_attr *qp_attr, 1752 int qp_attr_mask, 1753 struct ib_udata *udata); 1754 int (*query_qp)(struct ib_qp *qp, 1755 struct ib_qp_attr *qp_attr, 1756 int qp_attr_mask, 1757 struct ib_qp_init_attr *qp_init_attr); 1758 int (*destroy_qp)(struct ib_qp *qp); 1759 int (*post_send)(struct ib_qp *qp, 1760 struct ib_send_wr *send_wr, 1761 struct ib_send_wr **bad_send_wr); 1762 int (*post_recv)(struct ib_qp *qp, 1763 struct ib_recv_wr *recv_wr, 1764 struct ib_recv_wr **bad_recv_wr); 1765 struct ib_cq * (*create_cq)(struct ib_device *device, 1766 const struct ib_cq_init_attr *attr, 1767 struct ib_ucontext *context, 1768 struct ib_udata *udata); 1769 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, 1770 u16 cq_period); 1771 int (*destroy_cq)(struct ib_cq *cq); 1772 int (*resize_cq)(struct ib_cq *cq, int cqe, 1773 struct ib_udata *udata); 1774 int (*poll_cq)(struct ib_cq *cq, int num_entries, 1775 struct ib_wc *wc); 1776 int (*peek_cq)(struct ib_cq *cq, int wc_cnt); 1777 int (*req_notify_cq)(struct ib_cq *cq, 1778 enum ib_cq_notify_flags flags); 1779 int (*req_ncomp_notif)(struct ib_cq *cq, 1780 int wc_cnt); 1781 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd, 1782 int mr_access_flags); 1783 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd, 1784 u64 start, u64 length, 1785 u64 virt_addr, 1786 int mr_access_flags, 1787 struct ib_udata *udata); 1788 int (*rereg_user_mr)(struct ib_mr *mr, 1789 int flags, 1790 u64 start, u64 length, 1791 u64 virt_addr, 1792 int mr_access_flags, 1793 struct ib_pd *pd, 1794 struct ib_udata *udata); 1795 int (*dereg_mr)(struct ib_mr *mr); 1796 struct ib_mr * (*alloc_mr)(struct ib_pd *pd, 1797 enum ib_mr_type mr_type, 1798 u32 max_num_sg); 1799 int (*map_mr_sg)(struct ib_mr *mr, 1800 struct scatterlist *sg, 1801 int sg_nents); 1802 struct ib_mw * (*alloc_mw)(struct ib_pd *pd, 1803 enum ib_mw_type type); 1804 int (*dealloc_mw)(struct ib_mw *mw); 1805 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd, 1806 int mr_access_flags, 1807 struct ib_fmr_attr *fmr_attr); 1808 int (*map_phys_fmr)(struct ib_fmr *fmr, 1809 u64 *page_list, int list_len, 1810 u64 iova); 1811 int (*unmap_fmr)(struct list_head *fmr_list); 1812 int (*dealloc_fmr)(struct ib_fmr *fmr); 1813 int (*attach_mcast)(struct ib_qp *qp, 1814 union ib_gid *gid, 1815 u16 lid); 1816 int (*detach_mcast)(struct ib_qp *qp, 1817 union ib_gid *gid, 1818 u16 lid); 1819 int (*process_mad)(struct ib_device *device, 1820 int process_mad_flags, 1821 u8 port_num, 1822 const struct ib_wc *in_wc, 1823 const struct ib_grh *in_grh, 1824 const struct ib_mad_hdr *in_mad, 1825 size_t in_mad_size, 1826 struct ib_mad_hdr *out_mad, 1827 size_t *out_mad_size, 1828 u16 *out_mad_pkey_index); 1829 struct ib_xrcd * (*alloc_xrcd)(struct ib_device *device, 1830 struct ib_ucontext *ucontext, 1831 struct ib_udata *udata); 1832 int (*dealloc_xrcd)(struct ib_xrcd *xrcd); 1833 struct ib_flow * (*create_flow)(struct ib_qp *qp, 1834 struct ib_flow_attr 1835 *flow_attr, 1836 int domain); 1837 int (*destroy_flow)(struct ib_flow *flow_id); 1838 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask, 1839 struct ib_mr_status *mr_status); 1840 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext); 1841 1842 struct ib_dma_mapping_ops *dma_ops; 1843 1844 struct module *owner; 1845 struct device dev; 1846 struct kobject *ports_parent; 1847 struct list_head port_list; 1848 1849 enum { 1850 IB_DEV_UNINITIALIZED, 1851 IB_DEV_REGISTERED, 1852 IB_DEV_UNREGISTERED 1853 } reg_state; 1854 1855 int uverbs_abi_ver; 1856 u64 uverbs_cmd_mask; 1857 u64 uverbs_ex_cmd_mask; 1858 1859 char node_desc[64]; 1860 __be64 node_guid; 1861 u32 local_dma_lkey; 1862 u16 is_switch:1; 1863 u8 node_type; 1864 u8 phys_port_cnt; 1865 struct ib_device_attr attrs; 1866 1867 /** 1868 * The following mandatory functions are used only at device 1869 * registration. Keep functions such as these at the end of this 1870 * structure to avoid cache line misses when accessing struct ib_device 1871 * in fast paths. 1872 */ 1873 int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *); 1874 }; 1875 1876 struct ib_client { 1877 char *name; 1878 void (*add) (struct ib_device *); 1879 void (*remove)(struct ib_device *, void *client_data); 1880 1881 /* Returns the net_dev belonging to this ib_client and matching the 1882 * given parameters. 1883 * @dev: An RDMA device that the net_dev use for communication. 1884 * @port: A physical port number on the RDMA device. 1885 * @pkey: P_Key that the net_dev uses if applicable. 1886 * @gid: A GID that the net_dev uses to communicate. 1887 * @addr: An IP address the net_dev is configured with. 1888 * @client_data: The device's client data set by ib_set_client_data(). 1889 * 1890 * An ib_client that implements a net_dev on top of RDMA devices 1891 * (such as IP over IB) should implement this callback, allowing the 1892 * rdma_cm module to find the right net_dev for a given request. 1893 * 1894 * The caller is responsible for calling dev_put on the returned 1895 * netdev. */ 1896 struct net_device *(*get_net_dev_by_params)( 1897 struct ib_device *dev, 1898 u8 port, 1899 u16 pkey, 1900 const union ib_gid *gid, 1901 const struct sockaddr *addr, 1902 void *client_data); 1903 struct list_head list; 1904 }; 1905 1906 struct ib_device *ib_alloc_device(size_t size); 1907 void ib_dealloc_device(struct ib_device *device); 1908 1909 int ib_register_device(struct ib_device *device, 1910 int (*port_callback)(struct ib_device *, 1911 u8, struct kobject *)); 1912 void ib_unregister_device(struct ib_device *device); 1913 1914 int ib_register_client (struct ib_client *client); 1915 void ib_unregister_client(struct ib_client *client); 1916 1917 void *ib_get_client_data(struct ib_device *device, struct ib_client *client); 1918 void ib_set_client_data(struct ib_device *device, struct ib_client *client, 1919 void *data); 1920 1921 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len) 1922 { 1923 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0; 1924 } 1925 1926 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len) 1927 { 1928 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0; 1929 } 1930 1931 /** 1932 * ib_modify_qp_is_ok - Check that the supplied attribute mask 1933 * contains all required attributes and no attributes not allowed for 1934 * the given QP state transition. 1935 * @cur_state: Current QP state 1936 * @next_state: Next QP state 1937 * @type: QP type 1938 * @mask: Mask of supplied QP attributes 1939 * @ll : link layer of port 1940 * 1941 * This function is a helper function that a low-level driver's 1942 * modify_qp method can use to validate the consumer's input. It 1943 * checks that cur_state and next_state are valid QP states, that a 1944 * transition from cur_state to next_state is allowed by the IB spec, 1945 * and that the attribute mask supplied is allowed for the transition. 1946 */ 1947 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state, 1948 enum ib_qp_type type, enum ib_qp_attr_mask mask, 1949 enum rdma_link_layer ll); 1950 1951 int ib_register_event_handler (struct ib_event_handler *event_handler); 1952 int ib_unregister_event_handler(struct ib_event_handler *event_handler); 1953 void ib_dispatch_event(struct ib_event *event); 1954 1955 int ib_query_port(struct ib_device *device, 1956 u8 port_num, struct ib_port_attr *port_attr); 1957 1958 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device, 1959 u8 port_num); 1960 1961 /** 1962 * rdma_cap_ib_switch - Check if the device is IB switch 1963 * @device: Device to check 1964 * 1965 * Device driver is responsible for setting is_switch bit on 1966 * in ib_device structure at init time. 1967 * 1968 * Return: true if the device is IB switch. 1969 */ 1970 static inline bool rdma_cap_ib_switch(const struct ib_device *device) 1971 { 1972 return device->is_switch; 1973 } 1974 1975 /** 1976 * rdma_start_port - Return the first valid port number for the device 1977 * specified 1978 * 1979 * @device: Device to be checked 1980 * 1981 * Return start port number 1982 */ 1983 static inline u8 rdma_start_port(const struct ib_device *device) 1984 { 1985 return rdma_cap_ib_switch(device) ? 0 : 1; 1986 } 1987 1988 /** 1989 * rdma_end_port - Return the last valid port number for the device 1990 * specified 1991 * 1992 * @device: Device to be checked 1993 * 1994 * Return last port number 1995 */ 1996 static inline u8 rdma_end_port(const struct ib_device *device) 1997 { 1998 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt; 1999 } 2000 2001 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num) 2002 { 2003 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB; 2004 } 2005 2006 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num) 2007 { 2008 return device->port_immutable[port_num].core_cap_flags & 2009 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP); 2010 } 2011 2012 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num) 2013 { 2014 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP; 2015 } 2016 2017 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num) 2018 { 2019 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE; 2020 } 2021 2022 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num) 2023 { 2024 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP; 2025 } 2026 2027 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num) 2028 { 2029 return rdma_protocol_ib(device, port_num) || 2030 rdma_protocol_roce(device, port_num); 2031 } 2032 2033 /** 2034 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband 2035 * Management Datagrams. 2036 * @device: Device to check 2037 * @port_num: Port number to check 2038 * 2039 * Management Datagrams (MAD) are a required part of the InfiniBand 2040 * specification and are supported on all InfiniBand devices. A slightly 2041 * extended version are also supported on OPA interfaces. 2042 * 2043 * Return: true if the port supports sending/receiving of MAD packets. 2044 */ 2045 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num) 2046 { 2047 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD; 2048 } 2049 2050 /** 2051 * rdma_cap_opa_mad - Check if the port of device provides support for OPA 2052 * Management Datagrams. 2053 * @device: Device to check 2054 * @port_num: Port number to check 2055 * 2056 * Intel OmniPath devices extend and/or replace the InfiniBand Management 2057 * datagrams with their own versions. These OPA MADs share many but not all of 2058 * the characteristics of InfiniBand MADs. 2059 * 2060 * OPA MADs differ in the following ways: 2061 * 2062 * 1) MADs are variable size up to 2K 2063 * IBTA defined MADs remain fixed at 256 bytes 2064 * 2) OPA SMPs must carry valid PKeys 2065 * 3) OPA SMP packets are a different format 2066 * 2067 * Return: true if the port supports OPA MAD packet formats. 2068 */ 2069 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num) 2070 { 2071 return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD) 2072 == RDMA_CORE_CAP_OPA_MAD; 2073 } 2074 2075 /** 2076 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband 2077 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI). 2078 * @device: Device to check 2079 * @port_num: Port number to check 2080 * 2081 * Each InfiniBand node is required to provide a Subnet Management Agent 2082 * that the subnet manager can access. Prior to the fabric being fully 2083 * configured by the subnet manager, the SMA is accessed via a well known 2084 * interface called the Subnet Management Interface (SMI). This interface 2085 * uses directed route packets to communicate with the SM to get around the 2086 * chicken and egg problem of the SM needing to know what's on the fabric 2087 * in order to configure the fabric, and needing to configure the fabric in 2088 * order to send packets to the devices on the fabric. These directed 2089 * route packets do not need the fabric fully configured in order to reach 2090 * their destination. The SMI is the only method allowed to send 2091 * directed route packets on an InfiniBand fabric. 2092 * 2093 * Return: true if the port provides an SMI. 2094 */ 2095 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num) 2096 { 2097 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI; 2098 } 2099 2100 /** 2101 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband 2102 * Communication Manager. 2103 * @device: Device to check 2104 * @port_num: Port number to check 2105 * 2106 * The InfiniBand Communication Manager is one of many pre-defined General 2107 * Service Agents (GSA) that are accessed via the General Service 2108 * Interface (GSI). It's role is to facilitate establishment of connections 2109 * between nodes as well as other management related tasks for established 2110 * connections. 2111 * 2112 * Return: true if the port supports an IB CM (this does not guarantee that 2113 * a CM is actually running however). 2114 */ 2115 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num) 2116 { 2117 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM; 2118 } 2119 2120 /** 2121 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP 2122 * Communication Manager. 2123 * @device: Device to check 2124 * @port_num: Port number to check 2125 * 2126 * Similar to above, but specific to iWARP connections which have a different 2127 * managment protocol than InfiniBand. 2128 * 2129 * Return: true if the port supports an iWARP CM (this does not guarantee that 2130 * a CM is actually running however). 2131 */ 2132 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num) 2133 { 2134 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM; 2135 } 2136 2137 /** 2138 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband 2139 * Subnet Administration. 2140 * @device: Device to check 2141 * @port_num: Port number to check 2142 * 2143 * An InfiniBand Subnet Administration (SA) service is a pre-defined General 2144 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand 2145 * fabrics, devices should resolve routes to other hosts by contacting the 2146 * SA to query the proper route. 2147 * 2148 * Return: true if the port should act as a client to the fabric Subnet 2149 * Administration interface. This does not imply that the SA service is 2150 * running locally. 2151 */ 2152 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num) 2153 { 2154 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA; 2155 } 2156 2157 /** 2158 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband 2159 * Multicast. 2160 * @device: Device to check 2161 * @port_num: Port number to check 2162 * 2163 * InfiniBand multicast registration is more complex than normal IPv4 or 2164 * IPv6 multicast registration. Each Host Channel Adapter must register 2165 * with the Subnet Manager when it wishes to join a multicast group. It 2166 * should do so only once regardless of how many queue pairs it subscribes 2167 * to this group. And it should leave the group only after all queue pairs 2168 * attached to the group have been detached. 2169 * 2170 * Return: true if the port must undertake the additional adminstrative 2171 * overhead of registering/unregistering with the SM and tracking of the 2172 * total number of queue pairs attached to the multicast group. 2173 */ 2174 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num) 2175 { 2176 return rdma_cap_ib_sa(device, port_num); 2177 } 2178 2179 /** 2180 * rdma_cap_af_ib - Check if the port of device has the capability 2181 * Native Infiniband Address. 2182 * @device: Device to check 2183 * @port_num: Port number to check 2184 * 2185 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default 2186 * GID. RoCE uses a different mechanism, but still generates a GID via 2187 * a prescribed mechanism and port specific data. 2188 * 2189 * Return: true if the port uses a GID address to identify devices on the 2190 * network. 2191 */ 2192 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num) 2193 { 2194 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB; 2195 } 2196 2197 /** 2198 * rdma_cap_eth_ah - Check if the port of device has the capability 2199 * Ethernet Address Handle. 2200 * @device: Device to check 2201 * @port_num: Port number to check 2202 * 2203 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique 2204 * to fabricate GIDs over Ethernet/IP specific addresses native to the 2205 * port. Normally, packet headers are generated by the sending host 2206 * adapter, but when sending connectionless datagrams, we must manually 2207 * inject the proper headers for the fabric we are communicating over. 2208 * 2209 * Return: true if we are running as a RoCE port and must force the 2210 * addition of a Global Route Header built from our Ethernet Address 2211 * Handle into our header list for connectionless packets. 2212 */ 2213 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num) 2214 { 2215 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH; 2216 } 2217 2218 /** 2219 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port. 2220 * 2221 * @device: Device 2222 * @port_num: Port number 2223 * 2224 * This MAD size includes the MAD headers and MAD payload. No other headers 2225 * are included. 2226 * 2227 * Return the max MAD size required by the Port. Will return 0 if the port 2228 * does not support MADs 2229 */ 2230 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num) 2231 { 2232 return device->port_immutable[port_num].max_mad_size; 2233 } 2234 2235 /** 2236 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table 2237 * @device: Device to check 2238 * @port_num: Port number to check 2239 * 2240 * RoCE GID table mechanism manages the various GIDs for a device. 2241 * 2242 * NOTE: if allocating the port's GID table has failed, this call will still 2243 * return true, but any RoCE GID table API will fail. 2244 * 2245 * Return: true if the port uses RoCE GID table mechanism in order to manage 2246 * its GIDs. 2247 */ 2248 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device, 2249 u8 port_num) 2250 { 2251 return rdma_protocol_roce(device, port_num) && 2252 device->add_gid && device->del_gid; 2253 } 2254 2255 int ib_query_gid(struct ib_device *device, 2256 u8 port_num, int index, union ib_gid *gid, 2257 struct ib_gid_attr *attr); 2258 2259 int ib_query_pkey(struct ib_device *device, 2260 u8 port_num, u16 index, u16 *pkey); 2261 2262 int ib_modify_device(struct ib_device *device, 2263 int device_modify_mask, 2264 struct ib_device_modify *device_modify); 2265 2266 int ib_modify_port(struct ib_device *device, 2267 u8 port_num, int port_modify_mask, 2268 struct ib_port_modify *port_modify); 2269 2270 int ib_find_gid(struct ib_device *device, union ib_gid *gid, 2271 enum ib_gid_type gid_type, struct net_device *ndev, 2272 u8 *port_num, u16 *index); 2273 2274 int ib_find_pkey(struct ib_device *device, 2275 u8 port_num, u16 pkey, u16 *index); 2276 2277 struct ib_pd *ib_alloc_pd(struct ib_device *device); 2278 2279 void ib_dealloc_pd(struct ib_pd *pd); 2280 2281 /** 2282 * ib_create_ah - Creates an address handle for the given address vector. 2283 * @pd: The protection domain associated with the address handle. 2284 * @ah_attr: The attributes of the address vector. 2285 * 2286 * The address handle is used to reference a local or global destination 2287 * in all UD QP post sends. 2288 */ 2289 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr); 2290 2291 /** 2292 * ib_init_ah_from_wc - Initializes address handle attributes from a 2293 * work completion. 2294 * @device: Device on which the received message arrived. 2295 * @port_num: Port on which the received message arrived. 2296 * @wc: Work completion associated with the received message. 2297 * @grh: References the received global route header. This parameter is 2298 * ignored unless the work completion indicates that the GRH is valid. 2299 * @ah_attr: Returned attributes that can be used when creating an address 2300 * handle for replying to the message. 2301 */ 2302 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num, 2303 const struct ib_wc *wc, const struct ib_grh *grh, 2304 struct ib_ah_attr *ah_attr); 2305 2306 /** 2307 * ib_create_ah_from_wc - Creates an address handle associated with the 2308 * sender of the specified work completion. 2309 * @pd: The protection domain associated with the address handle. 2310 * @wc: Work completion information associated with a received message. 2311 * @grh: References the received global route header. This parameter is 2312 * ignored unless the work completion indicates that the GRH is valid. 2313 * @port_num: The outbound port number to associate with the address. 2314 * 2315 * The address handle is used to reference a local or global destination 2316 * in all UD QP post sends. 2317 */ 2318 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc, 2319 const struct ib_grh *grh, u8 port_num); 2320 2321 /** 2322 * ib_modify_ah - Modifies the address vector associated with an address 2323 * handle. 2324 * @ah: The address handle to modify. 2325 * @ah_attr: The new address vector attributes to associate with the 2326 * address handle. 2327 */ 2328 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr); 2329 2330 /** 2331 * ib_query_ah - Queries the address vector associated with an address 2332 * handle. 2333 * @ah: The address handle to query. 2334 * @ah_attr: The address vector attributes associated with the address 2335 * handle. 2336 */ 2337 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr); 2338 2339 /** 2340 * ib_destroy_ah - Destroys an address handle. 2341 * @ah: The address handle to destroy. 2342 */ 2343 int ib_destroy_ah(struct ib_ah *ah); 2344 2345 /** 2346 * ib_create_srq - Creates a SRQ associated with the specified protection 2347 * domain. 2348 * @pd: The protection domain associated with the SRQ. 2349 * @srq_init_attr: A list of initial attributes required to create the 2350 * SRQ. If SRQ creation succeeds, then the attributes are updated to 2351 * the actual capabilities of the created SRQ. 2352 * 2353 * srq_attr->max_wr and srq_attr->max_sge are read the determine the 2354 * requested size of the SRQ, and set to the actual values allocated 2355 * on return. If ib_create_srq() succeeds, then max_wr and max_sge 2356 * will always be at least as large as the requested values. 2357 */ 2358 struct ib_srq *ib_create_srq(struct ib_pd *pd, 2359 struct ib_srq_init_attr *srq_init_attr); 2360 2361 /** 2362 * ib_modify_srq - Modifies the attributes for the specified SRQ. 2363 * @srq: The SRQ to modify. 2364 * @srq_attr: On input, specifies the SRQ attributes to modify. On output, 2365 * the current values of selected SRQ attributes are returned. 2366 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ 2367 * are being modified. 2368 * 2369 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or 2370 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when 2371 * the number of receives queued drops below the limit. 2372 */ 2373 int ib_modify_srq(struct ib_srq *srq, 2374 struct ib_srq_attr *srq_attr, 2375 enum ib_srq_attr_mask srq_attr_mask); 2376 2377 /** 2378 * ib_query_srq - Returns the attribute list and current values for the 2379 * specified SRQ. 2380 * @srq: The SRQ to query. 2381 * @srq_attr: The attributes of the specified SRQ. 2382 */ 2383 int ib_query_srq(struct ib_srq *srq, 2384 struct ib_srq_attr *srq_attr); 2385 2386 /** 2387 * ib_destroy_srq - Destroys the specified SRQ. 2388 * @srq: The SRQ to destroy. 2389 */ 2390 int ib_destroy_srq(struct ib_srq *srq); 2391 2392 /** 2393 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ. 2394 * @srq: The SRQ to post the work request on. 2395 * @recv_wr: A list of work requests to post on the receive queue. 2396 * @bad_recv_wr: On an immediate failure, this parameter will reference 2397 * the work request that failed to be posted on the QP. 2398 */ 2399 static inline int ib_post_srq_recv(struct ib_srq *srq, 2400 struct ib_recv_wr *recv_wr, 2401 struct ib_recv_wr **bad_recv_wr) 2402 { 2403 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr); 2404 } 2405 2406 /** 2407 * ib_create_qp - Creates a QP associated with the specified protection 2408 * domain. 2409 * @pd: The protection domain associated with the QP. 2410 * @qp_init_attr: A list of initial attributes required to create the 2411 * QP. If QP creation succeeds, then the attributes are updated to 2412 * the actual capabilities of the created QP. 2413 */ 2414 struct ib_qp *ib_create_qp(struct ib_pd *pd, 2415 struct ib_qp_init_attr *qp_init_attr); 2416 2417 /** 2418 * ib_modify_qp - Modifies the attributes for the specified QP and then 2419 * transitions the QP to the given state. 2420 * @qp: The QP to modify. 2421 * @qp_attr: On input, specifies the QP attributes to modify. On output, 2422 * the current values of selected QP attributes are returned. 2423 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP 2424 * are being modified. 2425 */ 2426 int ib_modify_qp(struct ib_qp *qp, 2427 struct ib_qp_attr *qp_attr, 2428 int qp_attr_mask); 2429 2430 /** 2431 * ib_query_qp - Returns the attribute list and current values for the 2432 * specified QP. 2433 * @qp: The QP to query. 2434 * @qp_attr: The attributes of the specified QP. 2435 * @qp_attr_mask: A bit-mask used to select specific attributes to query. 2436 * @qp_init_attr: Additional attributes of the selected QP. 2437 * 2438 * The qp_attr_mask may be used to limit the query to gathering only the 2439 * selected attributes. 2440 */ 2441 int ib_query_qp(struct ib_qp *qp, 2442 struct ib_qp_attr *qp_attr, 2443 int qp_attr_mask, 2444 struct ib_qp_init_attr *qp_init_attr); 2445 2446 /** 2447 * ib_destroy_qp - Destroys the specified QP. 2448 * @qp: The QP to destroy. 2449 */ 2450 int ib_destroy_qp(struct ib_qp *qp); 2451 2452 /** 2453 * ib_open_qp - Obtain a reference to an existing sharable QP. 2454 * @xrcd - XRC domain 2455 * @qp_open_attr: Attributes identifying the QP to open. 2456 * 2457 * Returns a reference to a sharable QP. 2458 */ 2459 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd, 2460 struct ib_qp_open_attr *qp_open_attr); 2461 2462 /** 2463 * ib_close_qp - Release an external reference to a QP. 2464 * @qp: The QP handle to release 2465 * 2466 * The opened QP handle is released by the caller. The underlying 2467 * shared QP is not destroyed until all internal references are released. 2468 */ 2469 int ib_close_qp(struct ib_qp *qp); 2470 2471 /** 2472 * ib_post_send - Posts a list of work requests to the send queue of 2473 * the specified QP. 2474 * @qp: The QP to post the work request on. 2475 * @send_wr: A list of work requests to post on the send queue. 2476 * @bad_send_wr: On an immediate failure, this parameter will reference 2477 * the work request that failed to be posted on the QP. 2478 * 2479 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate 2480 * error is returned, the QP state shall not be affected, 2481 * ib_post_send() will return an immediate error after queueing any 2482 * earlier work requests in the list. 2483 */ 2484 static inline int ib_post_send(struct ib_qp *qp, 2485 struct ib_send_wr *send_wr, 2486 struct ib_send_wr **bad_send_wr) 2487 { 2488 return qp->device->post_send(qp, send_wr, bad_send_wr); 2489 } 2490 2491 /** 2492 * ib_post_recv - Posts a list of work requests to the receive queue of 2493 * the specified QP. 2494 * @qp: The QP to post the work request on. 2495 * @recv_wr: A list of work requests to post on the receive queue. 2496 * @bad_recv_wr: On an immediate failure, this parameter will reference 2497 * the work request that failed to be posted on the QP. 2498 */ 2499 static inline int ib_post_recv(struct ib_qp *qp, 2500 struct ib_recv_wr *recv_wr, 2501 struct ib_recv_wr **bad_recv_wr) 2502 { 2503 return qp->device->post_recv(qp, recv_wr, bad_recv_wr); 2504 } 2505 2506 struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private, 2507 int nr_cqe, int comp_vector, enum ib_poll_context poll_ctx); 2508 void ib_free_cq(struct ib_cq *cq); 2509 int ib_process_cq_direct(struct ib_cq *cq, int budget); 2510 2511 /** 2512 * ib_create_cq - Creates a CQ on the specified device. 2513 * @device: The device on which to create the CQ. 2514 * @comp_handler: A user-specified callback that is invoked when a 2515 * completion event occurs on the CQ. 2516 * @event_handler: A user-specified callback that is invoked when an 2517 * asynchronous event not associated with a completion occurs on the CQ. 2518 * @cq_context: Context associated with the CQ returned to the user via 2519 * the associated completion and event handlers. 2520 * @cq_attr: The attributes the CQ should be created upon. 2521 * 2522 * Users can examine the cq structure to determine the actual CQ size. 2523 */ 2524 struct ib_cq *ib_create_cq(struct ib_device *device, 2525 ib_comp_handler comp_handler, 2526 void (*event_handler)(struct ib_event *, void *), 2527 void *cq_context, 2528 const struct ib_cq_init_attr *cq_attr); 2529 2530 /** 2531 * ib_resize_cq - Modifies the capacity of the CQ. 2532 * @cq: The CQ to resize. 2533 * @cqe: The minimum size of the CQ. 2534 * 2535 * Users can examine the cq structure to determine the actual CQ size. 2536 */ 2537 int ib_resize_cq(struct ib_cq *cq, int cqe); 2538 2539 /** 2540 * ib_modify_cq - Modifies moderation params of the CQ 2541 * @cq: The CQ to modify. 2542 * @cq_count: number of CQEs that will trigger an event 2543 * @cq_period: max period of time in usec before triggering an event 2544 * 2545 */ 2546 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period); 2547 2548 /** 2549 * ib_destroy_cq - Destroys the specified CQ. 2550 * @cq: The CQ to destroy. 2551 */ 2552 int ib_destroy_cq(struct ib_cq *cq); 2553 2554 /** 2555 * ib_poll_cq - poll a CQ for completion(s) 2556 * @cq:the CQ being polled 2557 * @num_entries:maximum number of completions to return 2558 * @wc:array of at least @num_entries &struct ib_wc where completions 2559 * will be returned 2560 * 2561 * Poll a CQ for (possibly multiple) completions. If the return value 2562 * is < 0, an error occurred. If the return value is >= 0, it is the 2563 * number of completions returned. If the return value is 2564 * non-negative and < num_entries, then the CQ was emptied. 2565 */ 2566 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries, 2567 struct ib_wc *wc) 2568 { 2569 return cq->device->poll_cq(cq, num_entries, wc); 2570 } 2571 2572 /** 2573 * ib_peek_cq - Returns the number of unreaped completions currently 2574 * on the specified CQ. 2575 * @cq: The CQ to peek. 2576 * @wc_cnt: A minimum number of unreaped completions to check for. 2577 * 2578 * If the number of unreaped completions is greater than or equal to wc_cnt, 2579 * this function returns wc_cnt, otherwise, it returns the actual number of 2580 * unreaped completions. 2581 */ 2582 int ib_peek_cq(struct ib_cq *cq, int wc_cnt); 2583 2584 /** 2585 * ib_req_notify_cq - Request completion notification on a CQ. 2586 * @cq: The CQ to generate an event for. 2587 * @flags: 2588 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP 2589 * to request an event on the next solicited event or next work 2590 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS 2591 * may also be |ed in to request a hint about missed events, as 2592 * described below. 2593 * 2594 * Return Value: 2595 * < 0 means an error occurred while requesting notification 2596 * == 0 means notification was requested successfully, and if 2597 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events 2598 * were missed and it is safe to wait for another event. In 2599 * this case is it guaranteed that any work completions added 2600 * to the CQ since the last CQ poll will trigger a completion 2601 * notification event. 2602 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed 2603 * in. It means that the consumer must poll the CQ again to 2604 * make sure it is empty to avoid missing an event because of a 2605 * race between requesting notification and an entry being 2606 * added to the CQ. This return value means it is possible 2607 * (but not guaranteed) that a work completion has been added 2608 * to the CQ since the last poll without triggering a 2609 * completion notification event. 2610 */ 2611 static inline int ib_req_notify_cq(struct ib_cq *cq, 2612 enum ib_cq_notify_flags flags) 2613 { 2614 return cq->device->req_notify_cq(cq, flags); 2615 } 2616 2617 /** 2618 * ib_req_ncomp_notif - Request completion notification when there are 2619 * at least the specified number of unreaped completions on the CQ. 2620 * @cq: The CQ to generate an event for. 2621 * @wc_cnt: The number of unreaped completions that should be on the 2622 * CQ before an event is generated. 2623 */ 2624 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt) 2625 { 2626 return cq->device->req_ncomp_notif ? 2627 cq->device->req_ncomp_notif(cq, wc_cnt) : 2628 -ENOSYS; 2629 } 2630 2631 /** 2632 * ib_get_dma_mr - Returns a memory region for system memory that is 2633 * usable for DMA. 2634 * @pd: The protection domain associated with the memory region. 2635 * @mr_access_flags: Specifies the memory access rights. 2636 * 2637 * Note that the ib_dma_*() functions defined below must be used 2638 * to create/destroy addresses used with the Lkey or Rkey returned 2639 * by ib_get_dma_mr(). 2640 */ 2641 struct ib_mr *ib_get_dma_mr(struct ib_pd *pd, int mr_access_flags); 2642 2643 /** 2644 * ib_dma_mapping_error - check a DMA addr for error 2645 * @dev: The device for which the dma_addr was created 2646 * @dma_addr: The DMA address to check 2647 */ 2648 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr) 2649 { 2650 if (dev->dma_ops) 2651 return dev->dma_ops->mapping_error(dev, dma_addr); 2652 return dma_mapping_error(dev->dma_device, dma_addr); 2653 } 2654 2655 /** 2656 * ib_dma_map_single - Map a kernel virtual address to DMA address 2657 * @dev: The device for which the dma_addr is to be created 2658 * @cpu_addr: The kernel virtual address 2659 * @size: The size of the region in bytes 2660 * @direction: The direction of the DMA 2661 */ 2662 static inline u64 ib_dma_map_single(struct ib_device *dev, 2663 void *cpu_addr, size_t size, 2664 enum dma_data_direction direction) 2665 { 2666 if (dev->dma_ops) 2667 return dev->dma_ops->map_single(dev, cpu_addr, size, direction); 2668 return dma_map_single(dev->dma_device, cpu_addr, size, direction); 2669 } 2670 2671 /** 2672 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single() 2673 * @dev: The device for which the DMA address was created 2674 * @addr: The DMA address 2675 * @size: The size of the region in bytes 2676 * @direction: The direction of the DMA 2677 */ 2678 static inline void ib_dma_unmap_single(struct ib_device *dev, 2679 u64 addr, size_t size, 2680 enum dma_data_direction direction) 2681 { 2682 if (dev->dma_ops) 2683 dev->dma_ops->unmap_single(dev, addr, size, direction); 2684 else 2685 dma_unmap_single(dev->dma_device, addr, size, direction); 2686 } 2687 2688 static inline u64 ib_dma_map_single_attrs(struct ib_device *dev, 2689 void *cpu_addr, size_t size, 2690 enum dma_data_direction direction, 2691 struct dma_attrs *attrs) 2692 { 2693 return dma_map_single_attrs(dev->dma_device, cpu_addr, size, 2694 direction, attrs); 2695 } 2696 2697 static inline void ib_dma_unmap_single_attrs(struct ib_device *dev, 2698 u64 addr, size_t size, 2699 enum dma_data_direction direction, 2700 struct dma_attrs *attrs) 2701 { 2702 return dma_unmap_single_attrs(dev->dma_device, addr, size, 2703 direction, attrs); 2704 } 2705 2706 /** 2707 * ib_dma_map_page - Map a physical page to DMA address 2708 * @dev: The device for which the dma_addr is to be created 2709 * @page: The page to be mapped 2710 * @offset: The offset within the page 2711 * @size: The size of the region in bytes 2712 * @direction: The direction of the DMA 2713 */ 2714 static inline u64 ib_dma_map_page(struct ib_device *dev, 2715 struct page *page, 2716 unsigned long offset, 2717 size_t size, 2718 enum dma_data_direction direction) 2719 { 2720 if (dev->dma_ops) 2721 return dev->dma_ops->map_page(dev, page, offset, size, direction); 2722 return dma_map_page(dev->dma_device, page, offset, size, direction); 2723 } 2724 2725 /** 2726 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page() 2727 * @dev: The device for which the DMA address was created 2728 * @addr: The DMA address 2729 * @size: The size of the region in bytes 2730 * @direction: The direction of the DMA 2731 */ 2732 static inline void ib_dma_unmap_page(struct ib_device *dev, 2733 u64 addr, size_t size, 2734 enum dma_data_direction direction) 2735 { 2736 if (dev->dma_ops) 2737 dev->dma_ops->unmap_page(dev, addr, size, direction); 2738 else 2739 dma_unmap_page(dev->dma_device, addr, size, direction); 2740 } 2741 2742 /** 2743 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses 2744 * @dev: The device for which the DMA addresses are to be created 2745 * @sg: The array of scatter/gather entries 2746 * @nents: The number of scatter/gather entries 2747 * @direction: The direction of the DMA 2748 */ 2749 static inline int ib_dma_map_sg(struct ib_device *dev, 2750 struct scatterlist *sg, int nents, 2751 enum dma_data_direction direction) 2752 { 2753 if (dev->dma_ops) 2754 return dev->dma_ops->map_sg(dev, sg, nents, direction); 2755 return dma_map_sg(dev->dma_device, sg, nents, direction); 2756 } 2757 2758 /** 2759 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses 2760 * @dev: The device for which the DMA addresses were created 2761 * @sg: The array of scatter/gather entries 2762 * @nents: The number of scatter/gather entries 2763 * @direction: The direction of the DMA 2764 */ 2765 static inline void ib_dma_unmap_sg(struct ib_device *dev, 2766 struct scatterlist *sg, int nents, 2767 enum dma_data_direction direction) 2768 { 2769 if (dev->dma_ops) 2770 dev->dma_ops->unmap_sg(dev, sg, nents, direction); 2771 else 2772 dma_unmap_sg(dev->dma_device, sg, nents, direction); 2773 } 2774 2775 static inline int ib_dma_map_sg_attrs(struct ib_device *dev, 2776 struct scatterlist *sg, int nents, 2777 enum dma_data_direction direction, 2778 struct dma_attrs *attrs) 2779 { 2780 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, attrs); 2781 } 2782 2783 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev, 2784 struct scatterlist *sg, int nents, 2785 enum dma_data_direction direction, 2786 struct dma_attrs *attrs) 2787 { 2788 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, attrs); 2789 } 2790 /** 2791 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry 2792 * @dev: The device for which the DMA addresses were created 2793 * @sg: The scatter/gather entry 2794 * 2795 * Note: this function is obsolete. To do: change all occurrences of 2796 * ib_sg_dma_address() into sg_dma_address(). 2797 */ 2798 static inline u64 ib_sg_dma_address(struct ib_device *dev, 2799 struct scatterlist *sg) 2800 { 2801 return sg_dma_address(sg); 2802 } 2803 2804 /** 2805 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry 2806 * @dev: The device for which the DMA addresses were created 2807 * @sg: The scatter/gather entry 2808 * 2809 * Note: this function is obsolete. To do: change all occurrences of 2810 * ib_sg_dma_len() into sg_dma_len(). 2811 */ 2812 static inline unsigned int ib_sg_dma_len(struct ib_device *dev, 2813 struct scatterlist *sg) 2814 { 2815 return sg_dma_len(sg); 2816 } 2817 2818 /** 2819 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU 2820 * @dev: The device for which the DMA address was created 2821 * @addr: The DMA address 2822 * @size: The size of the region in bytes 2823 * @dir: The direction of the DMA 2824 */ 2825 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev, 2826 u64 addr, 2827 size_t size, 2828 enum dma_data_direction dir) 2829 { 2830 if (dev->dma_ops) 2831 dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir); 2832 else 2833 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir); 2834 } 2835 2836 /** 2837 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device 2838 * @dev: The device for which the DMA address was created 2839 * @addr: The DMA address 2840 * @size: The size of the region in bytes 2841 * @dir: The direction of the DMA 2842 */ 2843 static inline void ib_dma_sync_single_for_device(struct ib_device *dev, 2844 u64 addr, 2845 size_t size, 2846 enum dma_data_direction dir) 2847 { 2848 if (dev->dma_ops) 2849 dev->dma_ops->sync_single_for_device(dev, addr, size, dir); 2850 else 2851 dma_sync_single_for_device(dev->dma_device, addr, size, dir); 2852 } 2853 2854 /** 2855 * ib_dma_alloc_coherent - Allocate memory and map it for DMA 2856 * @dev: The device for which the DMA address is requested 2857 * @size: The size of the region to allocate in bytes 2858 * @dma_handle: A pointer for returning the DMA address of the region 2859 * @flag: memory allocator flags 2860 */ 2861 static inline void *ib_dma_alloc_coherent(struct ib_device *dev, 2862 size_t size, 2863 u64 *dma_handle, 2864 gfp_t flag) 2865 { 2866 if (dev->dma_ops) 2867 return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag); 2868 else { 2869 dma_addr_t handle; 2870 void *ret; 2871 2872 ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag); 2873 *dma_handle = handle; 2874 return ret; 2875 } 2876 } 2877 2878 /** 2879 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent() 2880 * @dev: The device for which the DMA addresses were allocated 2881 * @size: The size of the region 2882 * @cpu_addr: the address returned by ib_dma_alloc_coherent() 2883 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent() 2884 */ 2885 static inline void ib_dma_free_coherent(struct ib_device *dev, 2886 size_t size, void *cpu_addr, 2887 u64 dma_handle) 2888 { 2889 if (dev->dma_ops) 2890 dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle); 2891 else 2892 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle); 2893 } 2894 2895 /** 2896 * ib_dereg_mr - Deregisters a memory region and removes it from the 2897 * HCA translation table. 2898 * @mr: The memory region to deregister. 2899 * 2900 * This function can fail, if the memory region has memory windows bound to it. 2901 */ 2902 int ib_dereg_mr(struct ib_mr *mr); 2903 2904 struct ib_mr *ib_alloc_mr(struct ib_pd *pd, 2905 enum ib_mr_type mr_type, 2906 u32 max_num_sg); 2907 2908 /** 2909 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR 2910 * R_Key and L_Key. 2911 * @mr - struct ib_mr pointer to be updated. 2912 * @newkey - new key to be used. 2913 */ 2914 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey) 2915 { 2916 mr->lkey = (mr->lkey & 0xffffff00) | newkey; 2917 mr->rkey = (mr->rkey & 0xffffff00) | newkey; 2918 } 2919 2920 /** 2921 * ib_inc_rkey - increments the key portion of the given rkey. Can be used 2922 * for calculating a new rkey for type 2 memory windows. 2923 * @rkey - the rkey to increment. 2924 */ 2925 static inline u32 ib_inc_rkey(u32 rkey) 2926 { 2927 const u32 mask = 0x000000ff; 2928 return ((rkey + 1) & mask) | (rkey & ~mask); 2929 } 2930 2931 /** 2932 * ib_alloc_fmr - Allocates a unmapped fast memory region. 2933 * @pd: The protection domain associated with the unmapped region. 2934 * @mr_access_flags: Specifies the memory access rights. 2935 * @fmr_attr: Attributes of the unmapped region. 2936 * 2937 * A fast memory region must be mapped before it can be used as part of 2938 * a work request. 2939 */ 2940 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd, 2941 int mr_access_flags, 2942 struct ib_fmr_attr *fmr_attr); 2943 2944 /** 2945 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region. 2946 * @fmr: The fast memory region to associate with the pages. 2947 * @page_list: An array of physical pages to map to the fast memory region. 2948 * @list_len: The number of pages in page_list. 2949 * @iova: The I/O virtual address to use with the mapped region. 2950 */ 2951 static inline int ib_map_phys_fmr(struct ib_fmr *fmr, 2952 u64 *page_list, int list_len, 2953 u64 iova) 2954 { 2955 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova); 2956 } 2957 2958 /** 2959 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions. 2960 * @fmr_list: A linked list of fast memory regions to unmap. 2961 */ 2962 int ib_unmap_fmr(struct list_head *fmr_list); 2963 2964 /** 2965 * ib_dealloc_fmr - Deallocates a fast memory region. 2966 * @fmr: The fast memory region to deallocate. 2967 */ 2968 int ib_dealloc_fmr(struct ib_fmr *fmr); 2969 2970 /** 2971 * ib_attach_mcast - Attaches the specified QP to a multicast group. 2972 * @qp: QP to attach to the multicast group. The QP must be type 2973 * IB_QPT_UD. 2974 * @gid: Multicast group GID. 2975 * @lid: Multicast group LID in host byte order. 2976 * 2977 * In order to send and receive multicast packets, subnet 2978 * administration must have created the multicast group and configured 2979 * the fabric appropriately. The port associated with the specified 2980 * QP must also be a member of the multicast group. 2981 */ 2982 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid); 2983 2984 /** 2985 * ib_detach_mcast - Detaches the specified QP from a multicast group. 2986 * @qp: QP to detach from the multicast group. 2987 * @gid: Multicast group GID. 2988 * @lid: Multicast group LID in host byte order. 2989 */ 2990 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid); 2991 2992 /** 2993 * ib_alloc_xrcd - Allocates an XRC domain. 2994 * @device: The device on which to allocate the XRC domain. 2995 */ 2996 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device); 2997 2998 /** 2999 * ib_dealloc_xrcd - Deallocates an XRC domain. 3000 * @xrcd: The XRC domain to deallocate. 3001 */ 3002 int ib_dealloc_xrcd(struct ib_xrcd *xrcd); 3003 3004 struct ib_flow *ib_create_flow(struct ib_qp *qp, 3005 struct ib_flow_attr *flow_attr, int domain); 3006 int ib_destroy_flow(struct ib_flow *flow_id); 3007 3008 static inline int ib_check_mr_access(int flags) 3009 { 3010 /* 3011 * Local write permission is required if remote write or 3012 * remote atomic permission is also requested. 3013 */ 3014 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) && 3015 !(flags & IB_ACCESS_LOCAL_WRITE)) 3016 return -EINVAL; 3017 3018 return 0; 3019 } 3020 3021 /** 3022 * ib_check_mr_status: lightweight check of MR status. 3023 * This routine may provide status checks on a selected 3024 * ib_mr. first use is for signature status check. 3025 * 3026 * @mr: A memory region. 3027 * @check_mask: Bitmask of which checks to perform from 3028 * ib_mr_status_check enumeration. 3029 * @mr_status: The container of relevant status checks. 3030 * failed checks will be indicated in the status bitmask 3031 * and the relevant info shall be in the error item. 3032 */ 3033 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask, 3034 struct ib_mr_status *mr_status); 3035 3036 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port, 3037 u16 pkey, const union ib_gid *gid, 3038 const struct sockaddr *addr); 3039 3040 int ib_map_mr_sg(struct ib_mr *mr, 3041 struct scatterlist *sg, 3042 int sg_nents, 3043 unsigned int page_size); 3044 3045 static inline int 3046 ib_map_mr_sg_zbva(struct ib_mr *mr, 3047 struct scatterlist *sg, 3048 int sg_nents, 3049 unsigned int page_size) 3050 { 3051 int n; 3052 3053 n = ib_map_mr_sg(mr, sg, sg_nents, page_size); 3054 mr->iova = 0; 3055 3056 return n; 3057 } 3058 3059 int ib_sg_to_pages(struct ib_mr *mr, 3060 struct scatterlist *sgl, 3061 int sg_nents, 3062 int (*set_page)(struct ib_mr *, u64)); 3063 3064 #endif /* IB_VERBS_H */ 3065