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 <uapi/linux/if_ether.h> 53 54 #include <linux/atomic.h> 55 #include <linux/mmu_notifier.h> 56 #include <asm/uaccess.h> 57 58 extern struct workqueue_struct *ib_wq; 59 60 union ib_gid { 61 u8 raw[16]; 62 struct { 63 __be64 subnet_prefix; 64 __be64 interface_id; 65 } global; 66 }; 67 68 extern union ib_gid zgid; 69 70 struct ib_gid_attr { 71 struct net_device *ndev; 72 }; 73 74 enum rdma_node_type { 75 /* IB values map to NodeInfo:NodeType. */ 76 RDMA_NODE_IB_CA = 1, 77 RDMA_NODE_IB_SWITCH, 78 RDMA_NODE_IB_ROUTER, 79 RDMA_NODE_RNIC, 80 RDMA_NODE_USNIC, 81 RDMA_NODE_USNIC_UDP, 82 }; 83 84 enum rdma_transport_type { 85 RDMA_TRANSPORT_IB, 86 RDMA_TRANSPORT_IWARP, 87 RDMA_TRANSPORT_USNIC, 88 RDMA_TRANSPORT_USNIC_UDP 89 }; 90 91 enum rdma_protocol_type { 92 RDMA_PROTOCOL_IB, 93 RDMA_PROTOCOL_IBOE, 94 RDMA_PROTOCOL_IWARP, 95 RDMA_PROTOCOL_USNIC_UDP 96 }; 97 98 __attribute_const__ enum rdma_transport_type 99 rdma_node_get_transport(enum rdma_node_type node_type); 100 101 enum rdma_link_layer { 102 IB_LINK_LAYER_UNSPECIFIED, 103 IB_LINK_LAYER_INFINIBAND, 104 IB_LINK_LAYER_ETHERNET, 105 }; 106 107 enum ib_device_cap_flags { 108 IB_DEVICE_RESIZE_MAX_WR = 1, 109 IB_DEVICE_BAD_PKEY_CNTR = (1<<1), 110 IB_DEVICE_BAD_QKEY_CNTR = (1<<2), 111 IB_DEVICE_RAW_MULTI = (1<<3), 112 IB_DEVICE_AUTO_PATH_MIG = (1<<4), 113 IB_DEVICE_CHANGE_PHY_PORT = (1<<5), 114 IB_DEVICE_UD_AV_PORT_ENFORCE = (1<<6), 115 IB_DEVICE_CURR_QP_STATE_MOD = (1<<7), 116 IB_DEVICE_SHUTDOWN_PORT = (1<<8), 117 IB_DEVICE_INIT_TYPE = (1<<9), 118 IB_DEVICE_PORT_ACTIVE_EVENT = (1<<10), 119 IB_DEVICE_SYS_IMAGE_GUID = (1<<11), 120 IB_DEVICE_RC_RNR_NAK_GEN = (1<<12), 121 IB_DEVICE_SRQ_RESIZE = (1<<13), 122 IB_DEVICE_N_NOTIFY_CQ = (1<<14), 123 IB_DEVICE_LOCAL_DMA_LKEY = (1<<15), 124 IB_DEVICE_RESERVED = (1<<16), /* old SEND_W_INV */ 125 IB_DEVICE_MEM_WINDOW = (1<<17), 126 /* 127 * Devices should set IB_DEVICE_UD_IP_SUM if they support 128 * insertion of UDP and TCP checksum on outgoing UD IPoIB 129 * messages and can verify the validity of checksum for 130 * incoming messages. Setting this flag implies that the 131 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode. 132 */ 133 IB_DEVICE_UD_IP_CSUM = (1<<18), 134 IB_DEVICE_UD_TSO = (1<<19), 135 IB_DEVICE_XRC = (1<<20), 136 IB_DEVICE_MEM_MGT_EXTENSIONS = (1<<21), 137 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1<<22), 138 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1<<23), 139 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1<<24), 140 IB_DEVICE_RC_IP_CSUM = (1<<25), 141 IB_DEVICE_RAW_IP_CSUM = (1<<26), 142 IB_DEVICE_MANAGED_FLOW_STEERING = (1<<29), 143 IB_DEVICE_SIGNATURE_HANDOVER = (1<<30), 144 IB_DEVICE_ON_DEMAND_PAGING = (1<<31), 145 }; 146 147 enum ib_signature_prot_cap { 148 IB_PROT_T10DIF_TYPE_1 = 1, 149 IB_PROT_T10DIF_TYPE_2 = 1 << 1, 150 IB_PROT_T10DIF_TYPE_3 = 1 << 2, 151 }; 152 153 enum ib_signature_guard_cap { 154 IB_GUARD_T10DIF_CRC = 1, 155 IB_GUARD_T10DIF_CSUM = 1 << 1, 156 }; 157 158 enum ib_atomic_cap { 159 IB_ATOMIC_NONE, 160 IB_ATOMIC_HCA, 161 IB_ATOMIC_GLOB 162 }; 163 164 enum ib_odp_general_cap_bits { 165 IB_ODP_SUPPORT = 1 << 0, 166 }; 167 168 enum ib_odp_transport_cap_bits { 169 IB_ODP_SUPPORT_SEND = 1 << 0, 170 IB_ODP_SUPPORT_RECV = 1 << 1, 171 IB_ODP_SUPPORT_WRITE = 1 << 2, 172 IB_ODP_SUPPORT_READ = 1 << 3, 173 IB_ODP_SUPPORT_ATOMIC = 1 << 4, 174 }; 175 176 struct ib_odp_caps { 177 uint64_t general_caps; 178 struct { 179 uint32_t rc_odp_caps; 180 uint32_t uc_odp_caps; 181 uint32_t ud_odp_caps; 182 } per_transport_caps; 183 }; 184 185 enum ib_cq_creation_flags { 186 IB_CQ_FLAGS_TIMESTAMP_COMPLETION = 1 << 0, 187 }; 188 189 struct ib_cq_init_attr { 190 unsigned int cqe; 191 int comp_vector; 192 u32 flags; 193 }; 194 195 struct ib_device_attr { 196 u64 fw_ver; 197 __be64 sys_image_guid; 198 u64 max_mr_size; 199 u64 page_size_cap; 200 u32 vendor_id; 201 u32 vendor_part_id; 202 u32 hw_ver; 203 int max_qp; 204 int max_qp_wr; 205 int device_cap_flags; 206 int max_sge; 207 int max_sge_rd; 208 int max_cq; 209 int max_cqe; 210 int max_mr; 211 int max_pd; 212 int max_qp_rd_atom; 213 int max_ee_rd_atom; 214 int max_res_rd_atom; 215 int max_qp_init_rd_atom; 216 int max_ee_init_rd_atom; 217 enum ib_atomic_cap atomic_cap; 218 enum ib_atomic_cap masked_atomic_cap; 219 int max_ee; 220 int max_rdd; 221 int max_mw; 222 int max_raw_ipv6_qp; 223 int max_raw_ethy_qp; 224 int max_mcast_grp; 225 int max_mcast_qp_attach; 226 int max_total_mcast_qp_attach; 227 int max_ah; 228 int max_fmr; 229 int max_map_per_fmr; 230 int max_srq; 231 int max_srq_wr; 232 int max_srq_sge; 233 unsigned int max_fast_reg_page_list_len; 234 u16 max_pkeys; 235 u8 local_ca_ack_delay; 236 int sig_prot_cap; 237 int sig_guard_cap; 238 struct ib_odp_caps odp_caps; 239 uint64_t timestamp_mask; 240 uint64_t hca_core_clock; /* in KHZ */ 241 }; 242 243 enum ib_mtu { 244 IB_MTU_256 = 1, 245 IB_MTU_512 = 2, 246 IB_MTU_1024 = 3, 247 IB_MTU_2048 = 4, 248 IB_MTU_4096 = 5 249 }; 250 251 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu) 252 { 253 switch (mtu) { 254 case IB_MTU_256: return 256; 255 case IB_MTU_512: return 512; 256 case IB_MTU_1024: return 1024; 257 case IB_MTU_2048: return 2048; 258 case IB_MTU_4096: return 4096; 259 default: return -1; 260 } 261 } 262 263 enum ib_port_state { 264 IB_PORT_NOP = 0, 265 IB_PORT_DOWN = 1, 266 IB_PORT_INIT = 2, 267 IB_PORT_ARMED = 3, 268 IB_PORT_ACTIVE = 4, 269 IB_PORT_ACTIVE_DEFER = 5 270 }; 271 272 enum ib_port_cap_flags { 273 IB_PORT_SM = 1 << 1, 274 IB_PORT_NOTICE_SUP = 1 << 2, 275 IB_PORT_TRAP_SUP = 1 << 3, 276 IB_PORT_OPT_IPD_SUP = 1 << 4, 277 IB_PORT_AUTO_MIGR_SUP = 1 << 5, 278 IB_PORT_SL_MAP_SUP = 1 << 6, 279 IB_PORT_MKEY_NVRAM = 1 << 7, 280 IB_PORT_PKEY_NVRAM = 1 << 8, 281 IB_PORT_LED_INFO_SUP = 1 << 9, 282 IB_PORT_SM_DISABLED = 1 << 10, 283 IB_PORT_SYS_IMAGE_GUID_SUP = 1 << 11, 284 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP = 1 << 12, 285 IB_PORT_EXTENDED_SPEEDS_SUP = 1 << 14, 286 IB_PORT_CM_SUP = 1 << 16, 287 IB_PORT_SNMP_TUNNEL_SUP = 1 << 17, 288 IB_PORT_REINIT_SUP = 1 << 18, 289 IB_PORT_DEVICE_MGMT_SUP = 1 << 19, 290 IB_PORT_VENDOR_CLASS_SUP = 1 << 20, 291 IB_PORT_DR_NOTICE_SUP = 1 << 21, 292 IB_PORT_CAP_MASK_NOTICE_SUP = 1 << 22, 293 IB_PORT_BOOT_MGMT_SUP = 1 << 23, 294 IB_PORT_LINK_LATENCY_SUP = 1 << 24, 295 IB_PORT_CLIENT_REG_SUP = 1 << 25, 296 IB_PORT_IP_BASED_GIDS = 1 << 26, 297 }; 298 299 enum ib_port_width { 300 IB_WIDTH_1X = 1, 301 IB_WIDTH_4X = 2, 302 IB_WIDTH_8X = 4, 303 IB_WIDTH_12X = 8 304 }; 305 306 static inline int ib_width_enum_to_int(enum ib_port_width width) 307 { 308 switch (width) { 309 case IB_WIDTH_1X: return 1; 310 case IB_WIDTH_4X: return 4; 311 case IB_WIDTH_8X: return 8; 312 case IB_WIDTH_12X: return 12; 313 default: return -1; 314 } 315 } 316 317 enum ib_port_speed { 318 IB_SPEED_SDR = 1, 319 IB_SPEED_DDR = 2, 320 IB_SPEED_QDR = 4, 321 IB_SPEED_FDR10 = 8, 322 IB_SPEED_FDR = 16, 323 IB_SPEED_EDR = 32 324 }; 325 326 struct ib_protocol_stats { 327 /* TBD... */ 328 }; 329 330 struct iw_protocol_stats { 331 u64 ipInReceives; 332 u64 ipInHdrErrors; 333 u64 ipInTooBigErrors; 334 u64 ipInNoRoutes; 335 u64 ipInAddrErrors; 336 u64 ipInUnknownProtos; 337 u64 ipInTruncatedPkts; 338 u64 ipInDiscards; 339 u64 ipInDelivers; 340 u64 ipOutForwDatagrams; 341 u64 ipOutRequests; 342 u64 ipOutDiscards; 343 u64 ipOutNoRoutes; 344 u64 ipReasmTimeout; 345 u64 ipReasmReqds; 346 u64 ipReasmOKs; 347 u64 ipReasmFails; 348 u64 ipFragOKs; 349 u64 ipFragFails; 350 u64 ipFragCreates; 351 u64 ipInMcastPkts; 352 u64 ipOutMcastPkts; 353 u64 ipInBcastPkts; 354 u64 ipOutBcastPkts; 355 356 u64 tcpRtoAlgorithm; 357 u64 tcpRtoMin; 358 u64 tcpRtoMax; 359 u64 tcpMaxConn; 360 u64 tcpActiveOpens; 361 u64 tcpPassiveOpens; 362 u64 tcpAttemptFails; 363 u64 tcpEstabResets; 364 u64 tcpCurrEstab; 365 u64 tcpInSegs; 366 u64 tcpOutSegs; 367 u64 tcpRetransSegs; 368 u64 tcpInErrs; 369 u64 tcpOutRsts; 370 }; 371 372 union rdma_protocol_stats { 373 struct ib_protocol_stats ib; 374 struct iw_protocol_stats iw; 375 }; 376 377 /* Define bits for the various functionality this port needs to be supported by 378 * the core. 379 */ 380 /* Management 0x00000FFF */ 381 #define RDMA_CORE_CAP_IB_MAD 0x00000001 382 #define RDMA_CORE_CAP_IB_SMI 0x00000002 383 #define RDMA_CORE_CAP_IB_CM 0x00000004 384 #define RDMA_CORE_CAP_IW_CM 0x00000008 385 #define RDMA_CORE_CAP_IB_SA 0x00000010 386 #define RDMA_CORE_CAP_OPA_MAD 0x00000020 387 388 /* Address format 0x000FF000 */ 389 #define RDMA_CORE_CAP_AF_IB 0x00001000 390 #define RDMA_CORE_CAP_ETH_AH 0x00002000 391 392 /* Protocol 0xFFF00000 */ 393 #define RDMA_CORE_CAP_PROT_IB 0x00100000 394 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000 395 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000 396 397 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \ 398 | RDMA_CORE_CAP_IB_MAD \ 399 | RDMA_CORE_CAP_IB_SMI \ 400 | RDMA_CORE_CAP_IB_CM \ 401 | RDMA_CORE_CAP_IB_SA \ 402 | RDMA_CORE_CAP_AF_IB) 403 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \ 404 | RDMA_CORE_CAP_IB_MAD \ 405 | RDMA_CORE_CAP_IB_CM \ 406 | RDMA_CORE_CAP_AF_IB \ 407 | RDMA_CORE_CAP_ETH_AH) 408 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \ 409 | RDMA_CORE_CAP_IW_CM) 410 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \ 411 | RDMA_CORE_CAP_OPA_MAD) 412 413 struct ib_port_attr { 414 enum ib_port_state state; 415 enum ib_mtu max_mtu; 416 enum ib_mtu active_mtu; 417 int gid_tbl_len; 418 u32 port_cap_flags; 419 u32 max_msg_sz; 420 u32 bad_pkey_cntr; 421 u32 qkey_viol_cntr; 422 u16 pkey_tbl_len; 423 u16 lid; 424 u16 sm_lid; 425 u8 lmc; 426 u8 max_vl_num; 427 u8 sm_sl; 428 u8 subnet_timeout; 429 u8 init_type_reply; 430 u8 active_width; 431 u8 active_speed; 432 u8 phys_state; 433 }; 434 435 enum ib_device_modify_flags { 436 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0, 437 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1 438 }; 439 440 struct ib_device_modify { 441 u64 sys_image_guid; 442 char node_desc[64]; 443 }; 444 445 enum ib_port_modify_flags { 446 IB_PORT_SHUTDOWN = 1, 447 IB_PORT_INIT_TYPE = (1<<2), 448 IB_PORT_RESET_QKEY_CNTR = (1<<3) 449 }; 450 451 struct ib_port_modify { 452 u32 set_port_cap_mask; 453 u32 clr_port_cap_mask; 454 u8 init_type; 455 }; 456 457 enum ib_event_type { 458 IB_EVENT_CQ_ERR, 459 IB_EVENT_QP_FATAL, 460 IB_EVENT_QP_REQ_ERR, 461 IB_EVENT_QP_ACCESS_ERR, 462 IB_EVENT_COMM_EST, 463 IB_EVENT_SQ_DRAINED, 464 IB_EVENT_PATH_MIG, 465 IB_EVENT_PATH_MIG_ERR, 466 IB_EVENT_DEVICE_FATAL, 467 IB_EVENT_PORT_ACTIVE, 468 IB_EVENT_PORT_ERR, 469 IB_EVENT_LID_CHANGE, 470 IB_EVENT_PKEY_CHANGE, 471 IB_EVENT_SM_CHANGE, 472 IB_EVENT_SRQ_ERR, 473 IB_EVENT_SRQ_LIMIT_REACHED, 474 IB_EVENT_QP_LAST_WQE_REACHED, 475 IB_EVENT_CLIENT_REREGISTER, 476 IB_EVENT_GID_CHANGE, 477 }; 478 479 const char *__attribute_const__ ib_event_msg(enum ib_event_type event); 480 481 struct ib_event { 482 struct ib_device *device; 483 union { 484 struct ib_cq *cq; 485 struct ib_qp *qp; 486 struct ib_srq *srq; 487 u8 port_num; 488 } element; 489 enum ib_event_type event; 490 }; 491 492 struct ib_event_handler { 493 struct ib_device *device; 494 void (*handler)(struct ib_event_handler *, struct ib_event *); 495 struct list_head list; 496 }; 497 498 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \ 499 do { \ 500 (_ptr)->device = _device; \ 501 (_ptr)->handler = _handler; \ 502 INIT_LIST_HEAD(&(_ptr)->list); \ 503 } while (0) 504 505 struct ib_global_route { 506 union ib_gid dgid; 507 u32 flow_label; 508 u8 sgid_index; 509 u8 hop_limit; 510 u8 traffic_class; 511 }; 512 513 struct ib_grh { 514 __be32 version_tclass_flow; 515 __be16 paylen; 516 u8 next_hdr; 517 u8 hop_limit; 518 union ib_gid sgid; 519 union ib_gid dgid; 520 }; 521 522 enum { 523 IB_MULTICAST_QPN = 0xffffff 524 }; 525 526 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF) 527 528 enum ib_ah_flags { 529 IB_AH_GRH = 1 530 }; 531 532 enum ib_rate { 533 IB_RATE_PORT_CURRENT = 0, 534 IB_RATE_2_5_GBPS = 2, 535 IB_RATE_5_GBPS = 5, 536 IB_RATE_10_GBPS = 3, 537 IB_RATE_20_GBPS = 6, 538 IB_RATE_30_GBPS = 4, 539 IB_RATE_40_GBPS = 7, 540 IB_RATE_60_GBPS = 8, 541 IB_RATE_80_GBPS = 9, 542 IB_RATE_120_GBPS = 10, 543 IB_RATE_14_GBPS = 11, 544 IB_RATE_56_GBPS = 12, 545 IB_RATE_112_GBPS = 13, 546 IB_RATE_168_GBPS = 14, 547 IB_RATE_25_GBPS = 15, 548 IB_RATE_100_GBPS = 16, 549 IB_RATE_200_GBPS = 17, 550 IB_RATE_300_GBPS = 18 551 }; 552 553 /** 554 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the 555 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be 556 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec. 557 * @rate: rate to convert. 558 */ 559 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate); 560 561 /** 562 * ib_rate_to_mbps - Convert the IB rate enum to Mbps. 563 * For example, IB_RATE_2_5_GBPS will be converted to 2500. 564 * @rate: rate to convert. 565 */ 566 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate); 567 568 569 /** 570 * enum ib_mr_type - memory region type 571 * @IB_MR_TYPE_MEM_REG: memory region that is used for 572 * normal registration 573 * @IB_MR_TYPE_SIGNATURE: memory region that is used for 574 * signature operations (data-integrity 575 * capable regions) 576 */ 577 enum ib_mr_type { 578 IB_MR_TYPE_MEM_REG, 579 IB_MR_TYPE_SIGNATURE, 580 }; 581 582 /** 583 * Signature types 584 * IB_SIG_TYPE_NONE: Unprotected. 585 * IB_SIG_TYPE_T10_DIF: Type T10-DIF 586 */ 587 enum ib_signature_type { 588 IB_SIG_TYPE_NONE, 589 IB_SIG_TYPE_T10_DIF, 590 }; 591 592 /** 593 * Signature T10-DIF block-guard types 594 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules. 595 * IB_T10DIF_CSUM: Corresponds to IP checksum rules. 596 */ 597 enum ib_t10_dif_bg_type { 598 IB_T10DIF_CRC, 599 IB_T10DIF_CSUM 600 }; 601 602 /** 603 * struct ib_t10_dif_domain - Parameters specific for T10-DIF 604 * domain. 605 * @bg_type: T10-DIF block guard type (CRC|CSUM) 606 * @pi_interval: protection information interval. 607 * @bg: seed of guard computation. 608 * @app_tag: application tag of guard block 609 * @ref_tag: initial guard block reference tag. 610 * @ref_remap: Indicate wethear the reftag increments each block 611 * @app_escape: Indicate to skip block check if apptag=0xffff 612 * @ref_escape: Indicate to skip block check if reftag=0xffffffff 613 * @apptag_check_mask: check bitmask of application tag. 614 */ 615 struct ib_t10_dif_domain { 616 enum ib_t10_dif_bg_type bg_type; 617 u16 pi_interval; 618 u16 bg; 619 u16 app_tag; 620 u32 ref_tag; 621 bool ref_remap; 622 bool app_escape; 623 bool ref_escape; 624 u16 apptag_check_mask; 625 }; 626 627 /** 628 * struct ib_sig_domain - Parameters for signature domain 629 * @sig_type: specific signauture type 630 * @sig: union of all signature domain attributes that may 631 * be used to set domain layout. 632 */ 633 struct ib_sig_domain { 634 enum ib_signature_type sig_type; 635 union { 636 struct ib_t10_dif_domain dif; 637 } sig; 638 }; 639 640 /** 641 * struct ib_sig_attrs - Parameters for signature handover operation 642 * @check_mask: bitmask for signature byte check (8 bytes) 643 * @mem: memory domain layout desciptor. 644 * @wire: wire domain layout desciptor. 645 */ 646 struct ib_sig_attrs { 647 u8 check_mask; 648 struct ib_sig_domain mem; 649 struct ib_sig_domain wire; 650 }; 651 652 enum ib_sig_err_type { 653 IB_SIG_BAD_GUARD, 654 IB_SIG_BAD_REFTAG, 655 IB_SIG_BAD_APPTAG, 656 }; 657 658 /** 659 * struct ib_sig_err - signature error descriptor 660 */ 661 struct ib_sig_err { 662 enum ib_sig_err_type err_type; 663 u32 expected; 664 u32 actual; 665 u64 sig_err_offset; 666 u32 key; 667 }; 668 669 enum ib_mr_status_check { 670 IB_MR_CHECK_SIG_STATUS = 1, 671 }; 672 673 /** 674 * struct ib_mr_status - Memory region status container 675 * 676 * @fail_status: Bitmask of MR checks status. For each 677 * failed check a corresponding status bit is set. 678 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS 679 * failure. 680 */ 681 struct ib_mr_status { 682 u32 fail_status; 683 struct ib_sig_err sig_err; 684 }; 685 686 /** 687 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate 688 * enum. 689 * @mult: multiple to convert. 690 */ 691 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult); 692 693 struct ib_ah_attr { 694 struct ib_global_route grh; 695 u16 dlid; 696 u8 sl; 697 u8 src_path_bits; 698 u8 static_rate; 699 u8 ah_flags; 700 u8 port_num; 701 u8 dmac[ETH_ALEN]; 702 }; 703 704 enum ib_wc_status { 705 IB_WC_SUCCESS, 706 IB_WC_LOC_LEN_ERR, 707 IB_WC_LOC_QP_OP_ERR, 708 IB_WC_LOC_EEC_OP_ERR, 709 IB_WC_LOC_PROT_ERR, 710 IB_WC_WR_FLUSH_ERR, 711 IB_WC_MW_BIND_ERR, 712 IB_WC_BAD_RESP_ERR, 713 IB_WC_LOC_ACCESS_ERR, 714 IB_WC_REM_INV_REQ_ERR, 715 IB_WC_REM_ACCESS_ERR, 716 IB_WC_REM_OP_ERR, 717 IB_WC_RETRY_EXC_ERR, 718 IB_WC_RNR_RETRY_EXC_ERR, 719 IB_WC_LOC_RDD_VIOL_ERR, 720 IB_WC_REM_INV_RD_REQ_ERR, 721 IB_WC_REM_ABORT_ERR, 722 IB_WC_INV_EECN_ERR, 723 IB_WC_INV_EEC_STATE_ERR, 724 IB_WC_FATAL_ERR, 725 IB_WC_RESP_TIMEOUT_ERR, 726 IB_WC_GENERAL_ERR 727 }; 728 729 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status); 730 731 enum ib_wc_opcode { 732 IB_WC_SEND, 733 IB_WC_RDMA_WRITE, 734 IB_WC_RDMA_READ, 735 IB_WC_COMP_SWAP, 736 IB_WC_FETCH_ADD, 737 IB_WC_BIND_MW, 738 IB_WC_LSO, 739 IB_WC_LOCAL_INV, 740 IB_WC_REG_MR, 741 IB_WC_MASKED_COMP_SWAP, 742 IB_WC_MASKED_FETCH_ADD, 743 /* 744 * Set value of IB_WC_RECV so consumers can test if a completion is a 745 * receive by testing (opcode & IB_WC_RECV). 746 */ 747 IB_WC_RECV = 1 << 7, 748 IB_WC_RECV_RDMA_WITH_IMM 749 }; 750 751 enum ib_wc_flags { 752 IB_WC_GRH = 1, 753 IB_WC_WITH_IMM = (1<<1), 754 IB_WC_WITH_INVALIDATE = (1<<2), 755 IB_WC_IP_CSUM_OK = (1<<3), 756 IB_WC_WITH_SMAC = (1<<4), 757 IB_WC_WITH_VLAN = (1<<5), 758 }; 759 760 struct ib_wc { 761 u64 wr_id; 762 enum ib_wc_status status; 763 enum ib_wc_opcode opcode; 764 u32 vendor_err; 765 u32 byte_len; 766 struct ib_qp *qp; 767 union { 768 __be32 imm_data; 769 u32 invalidate_rkey; 770 } ex; 771 u32 src_qp; 772 int wc_flags; 773 u16 pkey_index; 774 u16 slid; 775 u8 sl; 776 u8 dlid_path_bits; 777 u8 port_num; /* valid only for DR SMPs on switches */ 778 u8 smac[ETH_ALEN]; 779 u16 vlan_id; 780 }; 781 782 enum ib_cq_notify_flags { 783 IB_CQ_SOLICITED = 1 << 0, 784 IB_CQ_NEXT_COMP = 1 << 1, 785 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP, 786 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2, 787 }; 788 789 enum ib_srq_type { 790 IB_SRQT_BASIC, 791 IB_SRQT_XRC 792 }; 793 794 enum ib_srq_attr_mask { 795 IB_SRQ_MAX_WR = 1 << 0, 796 IB_SRQ_LIMIT = 1 << 1, 797 }; 798 799 struct ib_srq_attr { 800 u32 max_wr; 801 u32 max_sge; 802 u32 srq_limit; 803 }; 804 805 struct ib_srq_init_attr { 806 void (*event_handler)(struct ib_event *, void *); 807 void *srq_context; 808 struct ib_srq_attr attr; 809 enum ib_srq_type srq_type; 810 811 union { 812 struct { 813 struct ib_xrcd *xrcd; 814 struct ib_cq *cq; 815 } xrc; 816 } ext; 817 }; 818 819 struct ib_qp_cap { 820 u32 max_send_wr; 821 u32 max_recv_wr; 822 u32 max_send_sge; 823 u32 max_recv_sge; 824 u32 max_inline_data; 825 }; 826 827 enum ib_sig_type { 828 IB_SIGNAL_ALL_WR, 829 IB_SIGNAL_REQ_WR 830 }; 831 832 enum ib_qp_type { 833 /* 834 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries 835 * here (and in that order) since the MAD layer uses them as 836 * indices into a 2-entry table. 837 */ 838 IB_QPT_SMI, 839 IB_QPT_GSI, 840 841 IB_QPT_RC, 842 IB_QPT_UC, 843 IB_QPT_UD, 844 IB_QPT_RAW_IPV6, 845 IB_QPT_RAW_ETHERTYPE, 846 IB_QPT_RAW_PACKET = 8, 847 IB_QPT_XRC_INI = 9, 848 IB_QPT_XRC_TGT, 849 IB_QPT_MAX, 850 /* Reserve a range for qp types internal to the low level driver. 851 * These qp types will not be visible at the IB core layer, so the 852 * IB_QPT_MAX usages should not be affected in the core layer 853 */ 854 IB_QPT_RESERVED1 = 0x1000, 855 IB_QPT_RESERVED2, 856 IB_QPT_RESERVED3, 857 IB_QPT_RESERVED4, 858 IB_QPT_RESERVED5, 859 IB_QPT_RESERVED6, 860 IB_QPT_RESERVED7, 861 IB_QPT_RESERVED8, 862 IB_QPT_RESERVED9, 863 IB_QPT_RESERVED10, 864 }; 865 866 enum ib_qp_create_flags { 867 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0, 868 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1, 869 IB_QP_CREATE_NETIF_QP = 1 << 5, 870 IB_QP_CREATE_SIGNATURE_EN = 1 << 6, 871 IB_QP_CREATE_USE_GFP_NOIO = 1 << 7, 872 /* reserve bits 26-31 for low level drivers' internal use */ 873 IB_QP_CREATE_RESERVED_START = 1 << 26, 874 IB_QP_CREATE_RESERVED_END = 1 << 31, 875 }; 876 877 /* 878 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler 879 * callback to destroy the passed in QP. 880 */ 881 882 struct ib_qp_init_attr { 883 void (*event_handler)(struct ib_event *, void *); 884 void *qp_context; 885 struct ib_cq *send_cq; 886 struct ib_cq *recv_cq; 887 struct ib_srq *srq; 888 struct ib_xrcd *xrcd; /* XRC TGT QPs only */ 889 struct ib_qp_cap cap; 890 enum ib_sig_type sq_sig_type; 891 enum ib_qp_type qp_type; 892 enum ib_qp_create_flags create_flags; 893 u8 port_num; /* special QP types only */ 894 }; 895 896 struct ib_qp_open_attr { 897 void (*event_handler)(struct ib_event *, void *); 898 void *qp_context; 899 u32 qp_num; 900 enum ib_qp_type qp_type; 901 }; 902 903 enum ib_rnr_timeout { 904 IB_RNR_TIMER_655_36 = 0, 905 IB_RNR_TIMER_000_01 = 1, 906 IB_RNR_TIMER_000_02 = 2, 907 IB_RNR_TIMER_000_03 = 3, 908 IB_RNR_TIMER_000_04 = 4, 909 IB_RNR_TIMER_000_06 = 5, 910 IB_RNR_TIMER_000_08 = 6, 911 IB_RNR_TIMER_000_12 = 7, 912 IB_RNR_TIMER_000_16 = 8, 913 IB_RNR_TIMER_000_24 = 9, 914 IB_RNR_TIMER_000_32 = 10, 915 IB_RNR_TIMER_000_48 = 11, 916 IB_RNR_TIMER_000_64 = 12, 917 IB_RNR_TIMER_000_96 = 13, 918 IB_RNR_TIMER_001_28 = 14, 919 IB_RNR_TIMER_001_92 = 15, 920 IB_RNR_TIMER_002_56 = 16, 921 IB_RNR_TIMER_003_84 = 17, 922 IB_RNR_TIMER_005_12 = 18, 923 IB_RNR_TIMER_007_68 = 19, 924 IB_RNR_TIMER_010_24 = 20, 925 IB_RNR_TIMER_015_36 = 21, 926 IB_RNR_TIMER_020_48 = 22, 927 IB_RNR_TIMER_030_72 = 23, 928 IB_RNR_TIMER_040_96 = 24, 929 IB_RNR_TIMER_061_44 = 25, 930 IB_RNR_TIMER_081_92 = 26, 931 IB_RNR_TIMER_122_88 = 27, 932 IB_RNR_TIMER_163_84 = 28, 933 IB_RNR_TIMER_245_76 = 29, 934 IB_RNR_TIMER_327_68 = 30, 935 IB_RNR_TIMER_491_52 = 31 936 }; 937 938 enum ib_qp_attr_mask { 939 IB_QP_STATE = 1, 940 IB_QP_CUR_STATE = (1<<1), 941 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2), 942 IB_QP_ACCESS_FLAGS = (1<<3), 943 IB_QP_PKEY_INDEX = (1<<4), 944 IB_QP_PORT = (1<<5), 945 IB_QP_QKEY = (1<<6), 946 IB_QP_AV = (1<<7), 947 IB_QP_PATH_MTU = (1<<8), 948 IB_QP_TIMEOUT = (1<<9), 949 IB_QP_RETRY_CNT = (1<<10), 950 IB_QP_RNR_RETRY = (1<<11), 951 IB_QP_RQ_PSN = (1<<12), 952 IB_QP_MAX_QP_RD_ATOMIC = (1<<13), 953 IB_QP_ALT_PATH = (1<<14), 954 IB_QP_MIN_RNR_TIMER = (1<<15), 955 IB_QP_SQ_PSN = (1<<16), 956 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17), 957 IB_QP_PATH_MIG_STATE = (1<<18), 958 IB_QP_CAP = (1<<19), 959 IB_QP_DEST_QPN = (1<<20), 960 IB_QP_RESERVED1 = (1<<21), 961 IB_QP_RESERVED2 = (1<<22), 962 IB_QP_RESERVED3 = (1<<23), 963 IB_QP_RESERVED4 = (1<<24), 964 }; 965 966 enum ib_qp_state { 967 IB_QPS_RESET, 968 IB_QPS_INIT, 969 IB_QPS_RTR, 970 IB_QPS_RTS, 971 IB_QPS_SQD, 972 IB_QPS_SQE, 973 IB_QPS_ERR 974 }; 975 976 enum ib_mig_state { 977 IB_MIG_MIGRATED, 978 IB_MIG_REARM, 979 IB_MIG_ARMED 980 }; 981 982 enum ib_mw_type { 983 IB_MW_TYPE_1 = 1, 984 IB_MW_TYPE_2 = 2 985 }; 986 987 struct ib_qp_attr { 988 enum ib_qp_state qp_state; 989 enum ib_qp_state cur_qp_state; 990 enum ib_mtu path_mtu; 991 enum ib_mig_state path_mig_state; 992 u32 qkey; 993 u32 rq_psn; 994 u32 sq_psn; 995 u32 dest_qp_num; 996 int qp_access_flags; 997 struct ib_qp_cap cap; 998 struct ib_ah_attr ah_attr; 999 struct ib_ah_attr alt_ah_attr; 1000 u16 pkey_index; 1001 u16 alt_pkey_index; 1002 u8 en_sqd_async_notify; 1003 u8 sq_draining; 1004 u8 max_rd_atomic; 1005 u8 max_dest_rd_atomic; 1006 u8 min_rnr_timer; 1007 u8 port_num; 1008 u8 timeout; 1009 u8 retry_cnt; 1010 u8 rnr_retry; 1011 u8 alt_port_num; 1012 u8 alt_timeout; 1013 }; 1014 1015 enum ib_wr_opcode { 1016 IB_WR_RDMA_WRITE, 1017 IB_WR_RDMA_WRITE_WITH_IMM, 1018 IB_WR_SEND, 1019 IB_WR_SEND_WITH_IMM, 1020 IB_WR_RDMA_READ, 1021 IB_WR_ATOMIC_CMP_AND_SWP, 1022 IB_WR_ATOMIC_FETCH_AND_ADD, 1023 IB_WR_LSO, 1024 IB_WR_SEND_WITH_INV, 1025 IB_WR_RDMA_READ_WITH_INV, 1026 IB_WR_LOCAL_INV, 1027 IB_WR_REG_MR, 1028 IB_WR_MASKED_ATOMIC_CMP_AND_SWP, 1029 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD, 1030 IB_WR_BIND_MW, 1031 IB_WR_REG_SIG_MR, 1032 /* reserve values for low level drivers' internal use. 1033 * These values will not be used at all in the ib core layer. 1034 */ 1035 IB_WR_RESERVED1 = 0xf0, 1036 IB_WR_RESERVED2, 1037 IB_WR_RESERVED3, 1038 IB_WR_RESERVED4, 1039 IB_WR_RESERVED5, 1040 IB_WR_RESERVED6, 1041 IB_WR_RESERVED7, 1042 IB_WR_RESERVED8, 1043 IB_WR_RESERVED9, 1044 IB_WR_RESERVED10, 1045 }; 1046 1047 enum ib_send_flags { 1048 IB_SEND_FENCE = 1, 1049 IB_SEND_SIGNALED = (1<<1), 1050 IB_SEND_SOLICITED = (1<<2), 1051 IB_SEND_INLINE = (1<<3), 1052 IB_SEND_IP_CSUM = (1<<4), 1053 1054 /* reserve bits 26-31 for low level drivers' internal use */ 1055 IB_SEND_RESERVED_START = (1 << 26), 1056 IB_SEND_RESERVED_END = (1 << 31), 1057 }; 1058 1059 struct ib_sge { 1060 u64 addr; 1061 u32 length; 1062 u32 lkey; 1063 }; 1064 1065 /** 1066 * struct ib_mw_bind_info - Parameters for a memory window bind operation. 1067 * @mr: A memory region to bind the memory window to. 1068 * @addr: The address where the memory window should begin. 1069 * @length: The length of the memory window, in bytes. 1070 * @mw_access_flags: Access flags from enum ib_access_flags for the window. 1071 * 1072 * This struct contains the shared parameters for type 1 and type 2 1073 * memory window bind operations. 1074 */ 1075 struct ib_mw_bind_info { 1076 struct ib_mr *mr; 1077 u64 addr; 1078 u64 length; 1079 int mw_access_flags; 1080 }; 1081 1082 struct ib_send_wr { 1083 struct ib_send_wr *next; 1084 u64 wr_id; 1085 struct ib_sge *sg_list; 1086 int num_sge; 1087 enum ib_wr_opcode opcode; 1088 int send_flags; 1089 union { 1090 __be32 imm_data; 1091 u32 invalidate_rkey; 1092 } ex; 1093 }; 1094 1095 struct ib_rdma_wr { 1096 struct ib_send_wr wr; 1097 u64 remote_addr; 1098 u32 rkey; 1099 }; 1100 1101 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr) 1102 { 1103 return container_of(wr, struct ib_rdma_wr, wr); 1104 } 1105 1106 struct ib_atomic_wr { 1107 struct ib_send_wr wr; 1108 u64 remote_addr; 1109 u64 compare_add; 1110 u64 swap; 1111 u64 compare_add_mask; 1112 u64 swap_mask; 1113 u32 rkey; 1114 }; 1115 1116 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr) 1117 { 1118 return container_of(wr, struct ib_atomic_wr, wr); 1119 } 1120 1121 struct ib_ud_wr { 1122 struct ib_send_wr wr; 1123 struct ib_ah *ah; 1124 void *header; 1125 int hlen; 1126 int mss; 1127 u32 remote_qpn; 1128 u32 remote_qkey; 1129 u16 pkey_index; /* valid for GSI only */ 1130 u8 port_num; /* valid for DR SMPs on switch only */ 1131 }; 1132 1133 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr) 1134 { 1135 return container_of(wr, struct ib_ud_wr, wr); 1136 } 1137 1138 struct ib_reg_wr { 1139 struct ib_send_wr wr; 1140 struct ib_mr *mr; 1141 u32 key; 1142 int access; 1143 }; 1144 1145 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr) 1146 { 1147 return container_of(wr, struct ib_reg_wr, wr); 1148 } 1149 1150 struct ib_bind_mw_wr { 1151 struct ib_send_wr wr; 1152 struct ib_mw *mw; 1153 /* The new rkey for the memory window. */ 1154 u32 rkey; 1155 struct ib_mw_bind_info bind_info; 1156 }; 1157 1158 static inline struct ib_bind_mw_wr *bind_mw_wr(struct ib_send_wr *wr) 1159 { 1160 return container_of(wr, struct ib_bind_mw_wr, wr); 1161 } 1162 1163 struct ib_sig_handover_wr { 1164 struct ib_send_wr wr; 1165 struct ib_sig_attrs *sig_attrs; 1166 struct ib_mr *sig_mr; 1167 int access_flags; 1168 struct ib_sge *prot; 1169 }; 1170 1171 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr) 1172 { 1173 return container_of(wr, struct ib_sig_handover_wr, wr); 1174 } 1175 1176 struct ib_recv_wr { 1177 struct ib_recv_wr *next; 1178 u64 wr_id; 1179 struct ib_sge *sg_list; 1180 int num_sge; 1181 }; 1182 1183 enum ib_access_flags { 1184 IB_ACCESS_LOCAL_WRITE = 1, 1185 IB_ACCESS_REMOTE_WRITE = (1<<1), 1186 IB_ACCESS_REMOTE_READ = (1<<2), 1187 IB_ACCESS_REMOTE_ATOMIC = (1<<3), 1188 IB_ACCESS_MW_BIND = (1<<4), 1189 IB_ZERO_BASED = (1<<5), 1190 IB_ACCESS_ON_DEMAND = (1<<6), 1191 }; 1192 1193 struct ib_phys_buf { 1194 u64 addr; 1195 u64 size; 1196 }; 1197 1198 struct ib_mr_attr { 1199 struct ib_pd *pd; 1200 u64 device_virt_addr; 1201 u64 size; 1202 int mr_access_flags; 1203 u32 lkey; 1204 u32 rkey; 1205 }; 1206 1207 enum ib_mr_rereg_flags { 1208 IB_MR_REREG_TRANS = 1, 1209 IB_MR_REREG_PD = (1<<1), 1210 IB_MR_REREG_ACCESS = (1<<2), 1211 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1) 1212 }; 1213 1214 /** 1215 * struct ib_mw_bind - Parameters for a type 1 memory window bind operation. 1216 * @wr_id: Work request id. 1217 * @send_flags: Flags from ib_send_flags enum. 1218 * @bind_info: More parameters of the bind operation. 1219 */ 1220 struct ib_mw_bind { 1221 u64 wr_id; 1222 int send_flags; 1223 struct ib_mw_bind_info bind_info; 1224 }; 1225 1226 struct ib_fmr_attr { 1227 int max_pages; 1228 int max_maps; 1229 u8 page_shift; 1230 }; 1231 1232 struct ib_umem; 1233 1234 struct ib_ucontext { 1235 struct ib_device *device; 1236 struct list_head pd_list; 1237 struct list_head mr_list; 1238 struct list_head mw_list; 1239 struct list_head cq_list; 1240 struct list_head qp_list; 1241 struct list_head srq_list; 1242 struct list_head ah_list; 1243 struct list_head xrcd_list; 1244 struct list_head rule_list; 1245 int closing; 1246 1247 struct pid *tgid; 1248 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING 1249 struct rb_root umem_tree; 1250 /* 1251 * Protects .umem_rbroot and tree, as well as odp_mrs_count and 1252 * mmu notifiers registration. 1253 */ 1254 struct rw_semaphore umem_rwsem; 1255 void (*invalidate_range)(struct ib_umem *umem, 1256 unsigned long start, unsigned long end); 1257 1258 struct mmu_notifier mn; 1259 atomic_t notifier_count; 1260 /* A list of umems that don't have private mmu notifier counters yet. */ 1261 struct list_head no_private_counters; 1262 int odp_mrs_count; 1263 #endif 1264 }; 1265 1266 struct ib_uobject { 1267 u64 user_handle; /* handle given to us by userspace */ 1268 struct ib_ucontext *context; /* associated user context */ 1269 void *object; /* containing object */ 1270 struct list_head list; /* link to context's list */ 1271 int id; /* index into kernel idr */ 1272 struct kref ref; 1273 struct rw_semaphore mutex; /* protects .live */ 1274 int live; 1275 }; 1276 1277 struct ib_udata { 1278 const void __user *inbuf; 1279 void __user *outbuf; 1280 size_t inlen; 1281 size_t outlen; 1282 }; 1283 1284 struct ib_pd { 1285 u32 local_dma_lkey; 1286 struct ib_device *device; 1287 struct ib_uobject *uobject; 1288 atomic_t usecnt; /* count all resources */ 1289 struct ib_mr *local_mr; 1290 }; 1291 1292 struct ib_xrcd { 1293 struct ib_device *device; 1294 atomic_t usecnt; /* count all exposed resources */ 1295 struct inode *inode; 1296 1297 struct mutex tgt_qp_mutex; 1298 struct list_head tgt_qp_list; 1299 }; 1300 1301 struct ib_ah { 1302 struct ib_device *device; 1303 struct ib_pd *pd; 1304 struct ib_uobject *uobject; 1305 }; 1306 1307 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context); 1308 1309 struct ib_cq { 1310 struct ib_device *device; 1311 struct ib_uobject *uobject; 1312 ib_comp_handler comp_handler; 1313 void (*event_handler)(struct ib_event *, void *); 1314 void *cq_context; 1315 int cqe; 1316 atomic_t usecnt; /* count number of work queues */ 1317 }; 1318 1319 struct ib_srq { 1320 struct ib_device *device; 1321 struct ib_pd *pd; 1322 struct ib_uobject *uobject; 1323 void (*event_handler)(struct ib_event *, void *); 1324 void *srq_context; 1325 enum ib_srq_type srq_type; 1326 atomic_t usecnt; 1327 1328 union { 1329 struct { 1330 struct ib_xrcd *xrcd; 1331 struct ib_cq *cq; 1332 u32 srq_num; 1333 } xrc; 1334 } ext; 1335 }; 1336 1337 struct ib_qp { 1338 struct ib_device *device; 1339 struct ib_pd *pd; 1340 struct ib_cq *send_cq; 1341 struct ib_cq *recv_cq; 1342 struct ib_srq *srq; 1343 struct ib_xrcd *xrcd; /* XRC TGT QPs only */ 1344 struct list_head xrcd_list; 1345 /* count times opened, mcast attaches, flow attaches */ 1346 atomic_t usecnt; 1347 struct list_head open_list; 1348 struct ib_qp *real_qp; 1349 struct ib_uobject *uobject; 1350 void (*event_handler)(struct ib_event *, void *); 1351 void *qp_context; 1352 u32 qp_num; 1353 enum ib_qp_type qp_type; 1354 }; 1355 1356 struct ib_mr { 1357 struct ib_device *device; 1358 struct ib_pd *pd; 1359 struct ib_uobject *uobject; 1360 u32 lkey; 1361 u32 rkey; 1362 u64 iova; 1363 u32 length; 1364 unsigned int page_size; 1365 atomic_t usecnt; /* count number of MWs */ 1366 }; 1367 1368 struct ib_mw { 1369 struct ib_device *device; 1370 struct ib_pd *pd; 1371 struct ib_uobject *uobject; 1372 u32 rkey; 1373 enum ib_mw_type type; 1374 }; 1375 1376 struct ib_fmr { 1377 struct ib_device *device; 1378 struct ib_pd *pd; 1379 struct list_head list; 1380 u32 lkey; 1381 u32 rkey; 1382 }; 1383 1384 /* Supported steering options */ 1385 enum ib_flow_attr_type { 1386 /* steering according to rule specifications */ 1387 IB_FLOW_ATTR_NORMAL = 0x0, 1388 /* default unicast and multicast rule - 1389 * receive all Eth traffic which isn't steered to any QP 1390 */ 1391 IB_FLOW_ATTR_ALL_DEFAULT = 0x1, 1392 /* default multicast rule - 1393 * receive all Eth multicast traffic which isn't steered to any QP 1394 */ 1395 IB_FLOW_ATTR_MC_DEFAULT = 0x2, 1396 /* sniffer rule - receive all port traffic */ 1397 IB_FLOW_ATTR_SNIFFER = 0x3 1398 }; 1399 1400 /* Supported steering header types */ 1401 enum ib_flow_spec_type { 1402 /* L2 headers*/ 1403 IB_FLOW_SPEC_ETH = 0x20, 1404 IB_FLOW_SPEC_IB = 0x22, 1405 /* L3 header*/ 1406 IB_FLOW_SPEC_IPV4 = 0x30, 1407 /* L4 headers*/ 1408 IB_FLOW_SPEC_TCP = 0x40, 1409 IB_FLOW_SPEC_UDP = 0x41 1410 }; 1411 #define IB_FLOW_SPEC_LAYER_MASK 0xF0 1412 #define IB_FLOW_SPEC_SUPPORT_LAYERS 4 1413 1414 /* Flow steering rule priority is set according to it's domain. 1415 * Lower domain value means higher priority. 1416 */ 1417 enum ib_flow_domain { 1418 IB_FLOW_DOMAIN_USER, 1419 IB_FLOW_DOMAIN_ETHTOOL, 1420 IB_FLOW_DOMAIN_RFS, 1421 IB_FLOW_DOMAIN_NIC, 1422 IB_FLOW_DOMAIN_NUM /* Must be last */ 1423 }; 1424 1425 struct ib_flow_eth_filter { 1426 u8 dst_mac[6]; 1427 u8 src_mac[6]; 1428 __be16 ether_type; 1429 __be16 vlan_tag; 1430 }; 1431 1432 struct ib_flow_spec_eth { 1433 enum ib_flow_spec_type type; 1434 u16 size; 1435 struct ib_flow_eth_filter val; 1436 struct ib_flow_eth_filter mask; 1437 }; 1438 1439 struct ib_flow_ib_filter { 1440 __be16 dlid; 1441 __u8 sl; 1442 }; 1443 1444 struct ib_flow_spec_ib { 1445 enum ib_flow_spec_type type; 1446 u16 size; 1447 struct ib_flow_ib_filter val; 1448 struct ib_flow_ib_filter mask; 1449 }; 1450 1451 struct ib_flow_ipv4_filter { 1452 __be32 src_ip; 1453 __be32 dst_ip; 1454 }; 1455 1456 struct ib_flow_spec_ipv4 { 1457 enum ib_flow_spec_type type; 1458 u16 size; 1459 struct ib_flow_ipv4_filter val; 1460 struct ib_flow_ipv4_filter mask; 1461 }; 1462 1463 struct ib_flow_tcp_udp_filter { 1464 __be16 dst_port; 1465 __be16 src_port; 1466 }; 1467 1468 struct ib_flow_spec_tcp_udp { 1469 enum ib_flow_spec_type type; 1470 u16 size; 1471 struct ib_flow_tcp_udp_filter val; 1472 struct ib_flow_tcp_udp_filter mask; 1473 }; 1474 1475 union ib_flow_spec { 1476 struct { 1477 enum ib_flow_spec_type type; 1478 u16 size; 1479 }; 1480 struct ib_flow_spec_eth eth; 1481 struct ib_flow_spec_ib ib; 1482 struct ib_flow_spec_ipv4 ipv4; 1483 struct ib_flow_spec_tcp_udp tcp_udp; 1484 }; 1485 1486 struct ib_flow_attr { 1487 enum ib_flow_attr_type type; 1488 u16 size; 1489 u16 priority; 1490 u32 flags; 1491 u8 num_of_specs; 1492 u8 port; 1493 /* Following are the optional layers according to user request 1494 * struct ib_flow_spec_xxx 1495 * struct ib_flow_spec_yyy 1496 */ 1497 }; 1498 1499 struct ib_flow { 1500 struct ib_qp *qp; 1501 struct ib_uobject *uobject; 1502 }; 1503 1504 struct ib_mad_hdr; 1505 struct ib_grh; 1506 1507 enum ib_process_mad_flags { 1508 IB_MAD_IGNORE_MKEY = 1, 1509 IB_MAD_IGNORE_BKEY = 2, 1510 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY 1511 }; 1512 1513 enum ib_mad_result { 1514 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */ 1515 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */ 1516 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */ 1517 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */ 1518 }; 1519 1520 #define IB_DEVICE_NAME_MAX 64 1521 1522 struct ib_cache { 1523 rwlock_t lock; 1524 struct ib_event_handler event_handler; 1525 struct ib_pkey_cache **pkey_cache; 1526 struct ib_gid_table **gid_cache; 1527 u8 *lmc_cache; 1528 }; 1529 1530 struct ib_dma_mapping_ops { 1531 int (*mapping_error)(struct ib_device *dev, 1532 u64 dma_addr); 1533 u64 (*map_single)(struct ib_device *dev, 1534 void *ptr, size_t size, 1535 enum dma_data_direction direction); 1536 void (*unmap_single)(struct ib_device *dev, 1537 u64 addr, size_t size, 1538 enum dma_data_direction direction); 1539 u64 (*map_page)(struct ib_device *dev, 1540 struct page *page, unsigned long offset, 1541 size_t size, 1542 enum dma_data_direction direction); 1543 void (*unmap_page)(struct ib_device *dev, 1544 u64 addr, size_t size, 1545 enum dma_data_direction direction); 1546 int (*map_sg)(struct ib_device *dev, 1547 struct scatterlist *sg, int nents, 1548 enum dma_data_direction direction); 1549 void (*unmap_sg)(struct ib_device *dev, 1550 struct scatterlist *sg, int nents, 1551 enum dma_data_direction direction); 1552 void (*sync_single_for_cpu)(struct ib_device *dev, 1553 u64 dma_handle, 1554 size_t size, 1555 enum dma_data_direction dir); 1556 void (*sync_single_for_device)(struct ib_device *dev, 1557 u64 dma_handle, 1558 size_t size, 1559 enum dma_data_direction dir); 1560 void *(*alloc_coherent)(struct ib_device *dev, 1561 size_t size, 1562 u64 *dma_handle, 1563 gfp_t flag); 1564 void (*free_coherent)(struct ib_device *dev, 1565 size_t size, void *cpu_addr, 1566 u64 dma_handle); 1567 }; 1568 1569 struct iw_cm_verbs; 1570 1571 struct ib_port_immutable { 1572 int pkey_tbl_len; 1573 int gid_tbl_len; 1574 u32 core_cap_flags; 1575 u32 max_mad_size; 1576 }; 1577 1578 struct ib_device { 1579 struct device *dma_device; 1580 1581 char name[IB_DEVICE_NAME_MAX]; 1582 1583 struct list_head event_handler_list; 1584 spinlock_t event_handler_lock; 1585 1586 spinlock_t client_data_lock; 1587 struct list_head core_list; 1588 /* Access to the client_data_list is protected by the client_data_lock 1589 * spinlock and the lists_rwsem read-write semaphore */ 1590 struct list_head client_data_list; 1591 1592 struct ib_cache cache; 1593 /** 1594 * port_immutable is indexed by port number 1595 */ 1596 struct ib_port_immutable *port_immutable; 1597 1598 int num_comp_vectors; 1599 1600 struct iw_cm_verbs *iwcm; 1601 1602 int (*get_protocol_stats)(struct ib_device *device, 1603 union rdma_protocol_stats *stats); 1604 int (*query_device)(struct ib_device *device, 1605 struct ib_device_attr *device_attr, 1606 struct ib_udata *udata); 1607 int (*query_port)(struct ib_device *device, 1608 u8 port_num, 1609 struct ib_port_attr *port_attr); 1610 enum rdma_link_layer (*get_link_layer)(struct ib_device *device, 1611 u8 port_num); 1612 /* When calling get_netdev, the HW vendor's driver should return the 1613 * net device of device @device at port @port_num or NULL if such 1614 * a net device doesn't exist. The vendor driver should call dev_hold 1615 * on this net device. The HW vendor's device driver must guarantee 1616 * that this function returns NULL before the net device reaches 1617 * NETDEV_UNREGISTER_FINAL state. 1618 */ 1619 struct net_device *(*get_netdev)(struct ib_device *device, 1620 u8 port_num); 1621 int (*query_gid)(struct ib_device *device, 1622 u8 port_num, int index, 1623 union ib_gid *gid); 1624 /* When calling add_gid, the HW vendor's driver should 1625 * add the gid of device @device at gid index @index of 1626 * port @port_num to be @gid. Meta-info of that gid (for example, 1627 * the network device related to this gid is available 1628 * at @attr. @context allows the HW vendor driver to store extra 1629 * information together with a GID entry. The HW vendor may allocate 1630 * memory to contain this information and store it in @context when a 1631 * new GID entry is written to. Params are consistent until the next 1632 * call of add_gid or delete_gid. The function should return 0 on 1633 * success or error otherwise. The function could be called 1634 * concurrently for different ports. This function is only called 1635 * when roce_gid_table is used. 1636 */ 1637 int (*add_gid)(struct ib_device *device, 1638 u8 port_num, 1639 unsigned int index, 1640 const union ib_gid *gid, 1641 const struct ib_gid_attr *attr, 1642 void **context); 1643 /* When calling del_gid, the HW vendor's driver should delete the 1644 * gid of device @device at gid index @index of port @port_num. 1645 * Upon the deletion of a GID entry, the HW vendor must free any 1646 * allocated memory. The caller will clear @context afterwards. 1647 * This function is only called when roce_gid_table is used. 1648 */ 1649 int (*del_gid)(struct ib_device *device, 1650 u8 port_num, 1651 unsigned int index, 1652 void **context); 1653 int (*query_pkey)(struct ib_device *device, 1654 u8 port_num, u16 index, u16 *pkey); 1655 int (*modify_device)(struct ib_device *device, 1656 int device_modify_mask, 1657 struct ib_device_modify *device_modify); 1658 int (*modify_port)(struct ib_device *device, 1659 u8 port_num, int port_modify_mask, 1660 struct ib_port_modify *port_modify); 1661 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device, 1662 struct ib_udata *udata); 1663 int (*dealloc_ucontext)(struct ib_ucontext *context); 1664 int (*mmap)(struct ib_ucontext *context, 1665 struct vm_area_struct *vma); 1666 struct ib_pd * (*alloc_pd)(struct ib_device *device, 1667 struct ib_ucontext *context, 1668 struct ib_udata *udata); 1669 int (*dealloc_pd)(struct ib_pd *pd); 1670 struct ib_ah * (*create_ah)(struct ib_pd *pd, 1671 struct ib_ah_attr *ah_attr); 1672 int (*modify_ah)(struct ib_ah *ah, 1673 struct ib_ah_attr *ah_attr); 1674 int (*query_ah)(struct ib_ah *ah, 1675 struct ib_ah_attr *ah_attr); 1676 int (*destroy_ah)(struct ib_ah *ah); 1677 struct ib_srq * (*create_srq)(struct ib_pd *pd, 1678 struct ib_srq_init_attr *srq_init_attr, 1679 struct ib_udata *udata); 1680 int (*modify_srq)(struct ib_srq *srq, 1681 struct ib_srq_attr *srq_attr, 1682 enum ib_srq_attr_mask srq_attr_mask, 1683 struct ib_udata *udata); 1684 int (*query_srq)(struct ib_srq *srq, 1685 struct ib_srq_attr *srq_attr); 1686 int (*destroy_srq)(struct ib_srq *srq); 1687 int (*post_srq_recv)(struct ib_srq *srq, 1688 struct ib_recv_wr *recv_wr, 1689 struct ib_recv_wr **bad_recv_wr); 1690 struct ib_qp * (*create_qp)(struct ib_pd *pd, 1691 struct ib_qp_init_attr *qp_init_attr, 1692 struct ib_udata *udata); 1693 int (*modify_qp)(struct ib_qp *qp, 1694 struct ib_qp_attr *qp_attr, 1695 int qp_attr_mask, 1696 struct ib_udata *udata); 1697 int (*query_qp)(struct ib_qp *qp, 1698 struct ib_qp_attr *qp_attr, 1699 int qp_attr_mask, 1700 struct ib_qp_init_attr *qp_init_attr); 1701 int (*destroy_qp)(struct ib_qp *qp); 1702 int (*post_send)(struct ib_qp *qp, 1703 struct ib_send_wr *send_wr, 1704 struct ib_send_wr **bad_send_wr); 1705 int (*post_recv)(struct ib_qp *qp, 1706 struct ib_recv_wr *recv_wr, 1707 struct ib_recv_wr **bad_recv_wr); 1708 struct ib_cq * (*create_cq)(struct ib_device *device, 1709 const struct ib_cq_init_attr *attr, 1710 struct ib_ucontext *context, 1711 struct ib_udata *udata); 1712 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, 1713 u16 cq_period); 1714 int (*destroy_cq)(struct ib_cq *cq); 1715 int (*resize_cq)(struct ib_cq *cq, int cqe, 1716 struct ib_udata *udata); 1717 int (*poll_cq)(struct ib_cq *cq, int num_entries, 1718 struct ib_wc *wc); 1719 int (*peek_cq)(struct ib_cq *cq, int wc_cnt); 1720 int (*req_notify_cq)(struct ib_cq *cq, 1721 enum ib_cq_notify_flags flags); 1722 int (*req_ncomp_notif)(struct ib_cq *cq, 1723 int wc_cnt); 1724 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd, 1725 int mr_access_flags); 1726 struct ib_mr * (*reg_phys_mr)(struct ib_pd *pd, 1727 struct ib_phys_buf *phys_buf_array, 1728 int num_phys_buf, 1729 int mr_access_flags, 1730 u64 *iova_start); 1731 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd, 1732 u64 start, u64 length, 1733 u64 virt_addr, 1734 int mr_access_flags, 1735 struct ib_udata *udata); 1736 int (*rereg_user_mr)(struct ib_mr *mr, 1737 int flags, 1738 u64 start, u64 length, 1739 u64 virt_addr, 1740 int mr_access_flags, 1741 struct ib_pd *pd, 1742 struct ib_udata *udata); 1743 int (*query_mr)(struct ib_mr *mr, 1744 struct ib_mr_attr *mr_attr); 1745 int (*dereg_mr)(struct ib_mr *mr); 1746 struct ib_mr * (*alloc_mr)(struct ib_pd *pd, 1747 enum ib_mr_type mr_type, 1748 u32 max_num_sg); 1749 int (*map_mr_sg)(struct ib_mr *mr, 1750 struct scatterlist *sg, 1751 int sg_nents); 1752 int (*rereg_phys_mr)(struct ib_mr *mr, 1753 int mr_rereg_mask, 1754 struct ib_pd *pd, 1755 struct ib_phys_buf *phys_buf_array, 1756 int num_phys_buf, 1757 int mr_access_flags, 1758 u64 *iova_start); 1759 struct ib_mw * (*alloc_mw)(struct ib_pd *pd, 1760 enum ib_mw_type type); 1761 int (*bind_mw)(struct ib_qp *qp, 1762 struct ib_mw *mw, 1763 struct ib_mw_bind *mw_bind); 1764 int (*dealloc_mw)(struct ib_mw *mw); 1765 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd, 1766 int mr_access_flags, 1767 struct ib_fmr_attr *fmr_attr); 1768 int (*map_phys_fmr)(struct ib_fmr *fmr, 1769 u64 *page_list, int list_len, 1770 u64 iova); 1771 int (*unmap_fmr)(struct list_head *fmr_list); 1772 int (*dealloc_fmr)(struct ib_fmr *fmr); 1773 int (*attach_mcast)(struct ib_qp *qp, 1774 union ib_gid *gid, 1775 u16 lid); 1776 int (*detach_mcast)(struct ib_qp *qp, 1777 union ib_gid *gid, 1778 u16 lid); 1779 int (*process_mad)(struct ib_device *device, 1780 int process_mad_flags, 1781 u8 port_num, 1782 const struct ib_wc *in_wc, 1783 const struct ib_grh *in_grh, 1784 const struct ib_mad_hdr *in_mad, 1785 size_t in_mad_size, 1786 struct ib_mad_hdr *out_mad, 1787 size_t *out_mad_size, 1788 u16 *out_mad_pkey_index); 1789 struct ib_xrcd * (*alloc_xrcd)(struct ib_device *device, 1790 struct ib_ucontext *ucontext, 1791 struct ib_udata *udata); 1792 int (*dealloc_xrcd)(struct ib_xrcd *xrcd); 1793 struct ib_flow * (*create_flow)(struct ib_qp *qp, 1794 struct ib_flow_attr 1795 *flow_attr, 1796 int domain); 1797 int (*destroy_flow)(struct ib_flow *flow_id); 1798 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask, 1799 struct ib_mr_status *mr_status); 1800 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext); 1801 1802 struct ib_dma_mapping_ops *dma_ops; 1803 1804 struct module *owner; 1805 struct device dev; 1806 struct kobject *ports_parent; 1807 struct list_head port_list; 1808 1809 enum { 1810 IB_DEV_UNINITIALIZED, 1811 IB_DEV_REGISTERED, 1812 IB_DEV_UNREGISTERED 1813 } reg_state; 1814 1815 int uverbs_abi_ver; 1816 u64 uverbs_cmd_mask; 1817 u64 uverbs_ex_cmd_mask; 1818 1819 char node_desc[64]; 1820 __be64 node_guid; 1821 u32 local_dma_lkey; 1822 u16 is_switch:1; 1823 u8 node_type; 1824 u8 phys_port_cnt; 1825 1826 /** 1827 * The following mandatory functions are used only at device 1828 * registration. Keep functions such as these at the end of this 1829 * structure to avoid cache line misses when accessing struct ib_device 1830 * in fast paths. 1831 */ 1832 int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *); 1833 }; 1834 1835 struct ib_client { 1836 char *name; 1837 void (*add) (struct ib_device *); 1838 void (*remove)(struct ib_device *, void *client_data); 1839 1840 /* Returns the net_dev belonging to this ib_client and matching the 1841 * given parameters. 1842 * @dev: An RDMA device that the net_dev use for communication. 1843 * @port: A physical port number on the RDMA device. 1844 * @pkey: P_Key that the net_dev uses if applicable. 1845 * @gid: A GID that the net_dev uses to communicate. 1846 * @addr: An IP address the net_dev is configured with. 1847 * @client_data: The device's client data set by ib_set_client_data(). 1848 * 1849 * An ib_client that implements a net_dev on top of RDMA devices 1850 * (such as IP over IB) should implement this callback, allowing the 1851 * rdma_cm module to find the right net_dev for a given request. 1852 * 1853 * The caller is responsible for calling dev_put on the returned 1854 * netdev. */ 1855 struct net_device *(*get_net_dev_by_params)( 1856 struct ib_device *dev, 1857 u8 port, 1858 u16 pkey, 1859 const union ib_gid *gid, 1860 const struct sockaddr *addr, 1861 void *client_data); 1862 struct list_head list; 1863 }; 1864 1865 struct ib_device *ib_alloc_device(size_t size); 1866 void ib_dealloc_device(struct ib_device *device); 1867 1868 int ib_register_device(struct ib_device *device, 1869 int (*port_callback)(struct ib_device *, 1870 u8, struct kobject *)); 1871 void ib_unregister_device(struct ib_device *device); 1872 1873 int ib_register_client (struct ib_client *client); 1874 void ib_unregister_client(struct ib_client *client); 1875 1876 void *ib_get_client_data(struct ib_device *device, struct ib_client *client); 1877 void ib_set_client_data(struct ib_device *device, struct ib_client *client, 1878 void *data); 1879 1880 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len) 1881 { 1882 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0; 1883 } 1884 1885 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len) 1886 { 1887 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0; 1888 } 1889 1890 /** 1891 * ib_modify_qp_is_ok - Check that the supplied attribute mask 1892 * contains all required attributes and no attributes not allowed for 1893 * the given QP state transition. 1894 * @cur_state: Current QP state 1895 * @next_state: Next QP state 1896 * @type: QP type 1897 * @mask: Mask of supplied QP attributes 1898 * @ll : link layer of port 1899 * 1900 * This function is a helper function that a low-level driver's 1901 * modify_qp method can use to validate the consumer's input. It 1902 * checks that cur_state and next_state are valid QP states, that a 1903 * transition from cur_state to next_state is allowed by the IB spec, 1904 * and that the attribute mask supplied is allowed for the transition. 1905 */ 1906 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state, 1907 enum ib_qp_type type, enum ib_qp_attr_mask mask, 1908 enum rdma_link_layer ll); 1909 1910 int ib_register_event_handler (struct ib_event_handler *event_handler); 1911 int ib_unregister_event_handler(struct ib_event_handler *event_handler); 1912 void ib_dispatch_event(struct ib_event *event); 1913 1914 int ib_query_device(struct ib_device *device, 1915 struct ib_device_attr *device_attr); 1916 1917 int ib_query_port(struct ib_device *device, 1918 u8 port_num, struct ib_port_attr *port_attr); 1919 1920 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device, 1921 u8 port_num); 1922 1923 /** 1924 * rdma_cap_ib_switch - Check if the device is IB switch 1925 * @device: Device to check 1926 * 1927 * Device driver is responsible for setting is_switch bit on 1928 * in ib_device structure at init time. 1929 * 1930 * Return: true if the device is IB switch. 1931 */ 1932 static inline bool rdma_cap_ib_switch(const struct ib_device *device) 1933 { 1934 return device->is_switch; 1935 } 1936 1937 /** 1938 * rdma_start_port - Return the first valid port number for the device 1939 * specified 1940 * 1941 * @device: Device to be checked 1942 * 1943 * Return start port number 1944 */ 1945 static inline u8 rdma_start_port(const struct ib_device *device) 1946 { 1947 return rdma_cap_ib_switch(device) ? 0 : 1; 1948 } 1949 1950 /** 1951 * rdma_end_port - Return the last valid port number for the device 1952 * specified 1953 * 1954 * @device: Device to be checked 1955 * 1956 * Return last port number 1957 */ 1958 static inline u8 rdma_end_port(const struct ib_device *device) 1959 { 1960 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt; 1961 } 1962 1963 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num) 1964 { 1965 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB; 1966 } 1967 1968 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num) 1969 { 1970 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE; 1971 } 1972 1973 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num) 1974 { 1975 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP; 1976 } 1977 1978 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num) 1979 { 1980 return device->port_immutable[port_num].core_cap_flags & 1981 (RDMA_CORE_CAP_PROT_IB | RDMA_CORE_CAP_PROT_ROCE); 1982 } 1983 1984 /** 1985 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband 1986 * Management Datagrams. 1987 * @device: Device to check 1988 * @port_num: Port number to check 1989 * 1990 * Management Datagrams (MAD) are a required part of the InfiniBand 1991 * specification and are supported on all InfiniBand devices. A slightly 1992 * extended version are also supported on OPA interfaces. 1993 * 1994 * Return: true if the port supports sending/receiving of MAD packets. 1995 */ 1996 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num) 1997 { 1998 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD; 1999 } 2000 2001 /** 2002 * rdma_cap_opa_mad - Check if the port of device provides support for OPA 2003 * Management Datagrams. 2004 * @device: Device to check 2005 * @port_num: Port number to check 2006 * 2007 * Intel OmniPath devices extend and/or replace the InfiniBand Management 2008 * datagrams with their own versions. These OPA MADs share many but not all of 2009 * the characteristics of InfiniBand MADs. 2010 * 2011 * OPA MADs differ in the following ways: 2012 * 2013 * 1) MADs are variable size up to 2K 2014 * IBTA defined MADs remain fixed at 256 bytes 2015 * 2) OPA SMPs must carry valid PKeys 2016 * 3) OPA SMP packets are a different format 2017 * 2018 * Return: true if the port supports OPA MAD packet formats. 2019 */ 2020 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num) 2021 { 2022 return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD) 2023 == RDMA_CORE_CAP_OPA_MAD; 2024 } 2025 2026 /** 2027 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband 2028 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI). 2029 * @device: Device to check 2030 * @port_num: Port number to check 2031 * 2032 * Each InfiniBand node is required to provide a Subnet Management Agent 2033 * that the subnet manager can access. Prior to the fabric being fully 2034 * configured by the subnet manager, the SMA is accessed via a well known 2035 * interface called the Subnet Management Interface (SMI). This interface 2036 * uses directed route packets to communicate with the SM to get around the 2037 * chicken and egg problem of the SM needing to know what's on the fabric 2038 * in order to configure the fabric, and needing to configure the fabric in 2039 * order to send packets to the devices on the fabric. These directed 2040 * route packets do not need the fabric fully configured in order to reach 2041 * their destination. The SMI is the only method allowed to send 2042 * directed route packets on an InfiniBand fabric. 2043 * 2044 * Return: true if the port provides an SMI. 2045 */ 2046 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num) 2047 { 2048 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI; 2049 } 2050 2051 /** 2052 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband 2053 * Communication Manager. 2054 * @device: Device to check 2055 * @port_num: Port number to check 2056 * 2057 * The InfiniBand Communication Manager is one of many pre-defined General 2058 * Service Agents (GSA) that are accessed via the General Service 2059 * Interface (GSI). It's role is to facilitate establishment of connections 2060 * between nodes as well as other management related tasks for established 2061 * connections. 2062 * 2063 * Return: true if the port supports an IB CM (this does not guarantee that 2064 * a CM is actually running however). 2065 */ 2066 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num) 2067 { 2068 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM; 2069 } 2070 2071 /** 2072 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP 2073 * Communication Manager. 2074 * @device: Device to check 2075 * @port_num: Port number to check 2076 * 2077 * Similar to above, but specific to iWARP connections which have a different 2078 * managment protocol than InfiniBand. 2079 * 2080 * Return: true if the port supports an iWARP CM (this does not guarantee that 2081 * a CM is actually running however). 2082 */ 2083 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num) 2084 { 2085 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM; 2086 } 2087 2088 /** 2089 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband 2090 * Subnet Administration. 2091 * @device: Device to check 2092 * @port_num: Port number to check 2093 * 2094 * An InfiniBand Subnet Administration (SA) service is a pre-defined General 2095 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand 2096 * fabrics, devices should resolve routes to other hosts by contacting the 2097 * SA to query the proper route. 2098 * 2099 * Return: true if the port should act as a client to the fabric Subnet 2100 * Administration interface. This does not imply that the SA service is 2101 * running locally. 2102 */ 2103 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num) 2104 { 2105 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA; 2106 } 2107 2108 /** 2109 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband 2110 * Multicast. 2111 * @device: Device to check 2112 * @port_num: Port number to check 2113 * 2114 * InfiniBand multicast registration is more complex than normal IPv4 or 2115 * IPv6 multicast registration. Each Host Channel Adapter must register 2116 * with the Subnet Manager when it wishes to join a multicast group. It 2117 * should do so only once regardless of how many queue pairs it subscribes 2118 * to this group. And it should leave the group only after all queue pairs 2119 * attached to the group have been detached. 2120 * 2121 * Return: true if the port must undertake the additional adminstrative 2122 * overhead of registering/unregistering with the SM and tracking of the 2123 * total number of queue pairs attached to the multicast group. 2124 */ 2125 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num) 2126 { 2127 return rdma_cap_ib_sa(device, port_num); 2128 } 2129 2130 /** 2131 * rdma_cap_af_ib - Check if the port of device has the capability 2132 * Native Infiniband Address. 2133 * @device: Device to check 2134 * @port_num: Port number to check 2135 * 2136 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default 2137 * GID. RoCE uses a different mechanism, but still generates a GID via 2138 * a prescribed mechanism and port specific data. 2139 * 2140 * Return: true if the port uses a GID address to identify devices on the 2141 * network. 2142 */ 2143 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num) 2144 { 2145 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB; 2146 } 2147 2148 /** 2149 * rdma_cap_eth_ah - Check if the port of device has the capability 2150 * Ethernet Address Handle. 2151 * @device: Device to check 2152 * @port_num: Port number to check 2153 * 2154 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique 2155 * to fabricate GIDs over Ethernet/IP specific addresses native to the 2156 * port. Normally, packet headers are generated by the sending host 2157 * adapter, but when sending connectionless datagrams, we must manually 2158 * inject the proper headers for the fabric we are communicating over. 2159 * 2160 * Return: true if we are running as a RoCE port and must force the 2161 * addition of a Global Route Header built from our Ethernet Address 2162 * Handle into our header list for connectionless packets. 2163 */ 2164 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num) 2165 { 2166 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH; 2167 } 2168 2169 /** 2170 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port. 2171 * 2172 * @device: Device 2173 * @port_num: Port number 2174 * 2175 * This MAD size includes the MAD headers and MAD payload. No other headers 2176 * are included. 2177 * 2178 * Return the max MAD size required by the Port. Will return 0 if the port 2179 * does not support MADs 2180 */ 2181 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num) 2182 { 2183 return device->port_immutable[port_num].max_mad_size; 2184 } 2185 2186 /** 2187 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table 2188 * @device: Device to check 2189 * @port_num: Port number to check 2190 * 2191 * RoCE GID table mechanism manages the various GIDs for a device. 2192 * 2193 * NOTE: if allocating the port's GID table has failed, this call will still 2194 * return true, but any RoCE GID table API will fail. 2195 * 2196 * Return: true if the port uses RoCE GID table mechanism in order to manage 2197 * its GIDs. 2198 */ 2199 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device, 2200 u8 port_num) 2201 { 2202 return rdma_protocol_roce(device, port_num) && 2203 device->add_gid && device->del_gid; 2204 } 2205 2206 int ib_query_gid(struct ib_device *device, 2207 u8 port_num, int index, union ib_gid *gid, 2208 struct ib_gid_attr *attr); 2209 2210 int ib_query_pkey(struct ib_device *device, 2211 u8 port_num, u16 index, u16 *pkey); 2212 2213 int ib_modify_device(struct ib_device *device, 2214 int device_modify_mask, 2215 struct ib_device_modify *device_modify); 2216 2217 int ib_modify_port(struct ib_device *device, 2218 u8 port_num, int port_modify_mask, 2219 struct ib_port_modify *port_modify); 2220 2221 int ib_find_gid(struct ib_device *device, union ib_gid *gid, 2222 struct net_device *ndev, u8 *port_num, u16 *index); 2223 2224 int ib_find_pkey(struct ib_device *device, 2225 u8 port_num, u16 pkey, u16 *index); 2226 2227 struct ib_pd *ib_alloc_pd(struct ib_device *device); 2228 2229 void ib_dealloc_pd(struct ib_pd *pd); 2230 2231 /** 2232 * ib_create_ah - Creates an address handle for the given address vector. 2233 * @pd: The protection domain associated with the address handle. 2234 * @ah_attr: The attributes of the address vector. 2235 * 2236 * The address handle is used to reference a local or global destination 2237 * in all UD QP post sends. 2238 */ 2239 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr); 2240 2241 /** 2242 * ib_init_ah_from_wc - Initializes address handle attributes from a 2243 * work completion. 2244 * @device: Device on which the received message arrived. 2245 * @port_num: Port on which the received message arrived. 2246 * @wc: Work completion associated with the received message. 2247 * @grh: References the received global route header. This parameter is 2248 * ignored unless the work completion indicates that the GRH is valid. 2249 * @ah_attr: Returned attributes that can be used when creating an address 2250 * handle for replying to the message. 2251 */ 2252 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num, 2253 const struct ib_wc *wc, const struct ib_grh *grh, 2254 struct ib_ah_attr *ah_attr); 2255 2256 /** 2257 * ib_create_ah_from_wc - Creates an address handle associated with the 2258 * sender of the specified work completion. 2259 * @pd: The protection domain associated with the address handle. 2260 * @wc: Work completion information associated with a received message. 2261 * @grh: References the received global route header. This parameter is 2262 * ignored unless the work completion indicates that the GRH is valid. 2263 * @port_num: The outbound port number to associate with the address. 2264 * 2265 * The address handle is used to reference a local or global destination 2266 * in all UD QP post sends. 2267 */ 2268 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc, 2269 const struct ib_grh *grh, u8 port_num); 2270 2271 /** 2272 * ib_modify_ah - Modifies the address vector associated with an address 2273 * handle. 2274 * @ah: The address handle to modify. 2275 * @ah_attr: The new address vector attributes to associate with the 2276 * address handle. 2277 */ 2278 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr); 2279 2280 /** 2281 * ib_query_ah - Queries the address vector associated with an address 2282 * handle. 2283 * @ah: The address handle to query. 2284 * @ah_attr: The address vector attributes associated with the address 2285 * handle. 2286 */ 2287 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr); 2288 2289 /** 2290 * ib_destroy_ah - Destroys an address handle. 2291 * @ah: The address handle to destroy. 2292 */ 2293 int ib_destroy_ah(struct ib_ah *ah); 2294 2295 /** 2296 * ib_create_srq - Creates a SRQ associated with the specified protection 2297 * domain. 2298 * @pd: The protection domain associated with the SRQ. 2299 * @srq_init_attr: A list of initial attributes required to create the 2300 * SRQ. If SRQ creation succeeds, then the attributes are updated to 2301 * the actual capabilities of the created SRQ. 2302 * 2303 * srq_attr->max_wr and srq_attr->max_sge are read the determine the 2304 * requested size of the SRQ, and set to the actual values allocated 2305 * on return. If ib_create_srq() succeeds, then max_wr and max_sge 2306 * will always be at least as large as the requested values. 2307 */ 2308 struct ib_srq *ib_create_srq(struct ib_pd *pd, 2309 struct ib_srq_init_attr *srq_init_attr); 2310 2311 /** 2312 * ib_modify_srq - Modifies the attributes for the specified SRQ. 2313 * @srq: The SRQ to modify. 2314 * @srq_attr: On input, specifies the SRQ attributes to modify. On output, 2315 * the current values of selected SRQ attributes are returned. 2316 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ 2317 * are being modified. 2318 * 2319 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or 2320 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when 2321 * the number of receives queued drops below the limit. 2322 */ 2323 int ib_modify_srq(struct ib_srq *srq, 2324 struct ib_srq_attr *srq_attr, 2325 enum ib_srq_attr_mask srq_attr_mask); 2326 2327 /** 2328 * ib_query_srq - Returns the attribute list and current values for the 2329 * specified SRQ. 2330 * @srq: The SRQ to query. 2331 * @srq_attr: The attributes of the specified SRQ. 2332 */ 2333 int ib_query_srq(struct ib_srq *srq, 2334 struct ib_srq_attr *srq_attr); 2335 2336 /** 2337 * ib_destroy_srq - Destroys the specified SRQ. 2338 * @srq: The SRQ to destroy. 2339 */ 2340 int ib_destroy_srq(struct ib_srq *srq); 2341 2342 /** 2343 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ. 2344 * @srq: The SRQ to post the work request on. 2345 * @recv_wr: A list of work requests to post on the receive queue. 2346 * @bad_recv_wr: On an immediate failure, this parameter will reference 2347 * the work request that failed to be posted on the QP. 2348 */ 2349 static inline int ib_post_srq_recv(struct ib_srq *srq, 2350 struct ib_recv_wr *recv_wr, 2351 struct ib_recv_wr **bad_recv_wr) 2352 { 2353 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr); 2354 } 2355 2356 /** 2357 * ib_create_qp - Creates a QP associated with the specified protection 2358 * domain. 2359 * @pd: The protection domain associated with the QP. 2360 * @qp_init_attr: A list of initial attributes required to create the 2361 * QP. If QP creation succeeds, then the attributes are updated to 2362 * the actual capabilities of the created QP. 2363 */ 2364 struct ib_qp *ib_create_qp(struct ib_pd *pd, 2365 struct ib_qp_init_attr *qp_init_attr); 2366 2367 /** 2368 * ib_modify_qp - Modifies the attributes for the specified QP and then 2369 * transitions the QP to the given state. 2370 * @qp: The QP to modify. 2371 * @qp_attr: On input, specifies the QP attributes to modify. On output, 2372 * the current values of selected QP attributes are returned. 2373 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP 2374 * are being modified. 2375 */ 2376 int ib_modify_qp(struct ib_qp *qp, 2377 struct ib_qp_attr *qp_attr, 2378 int qp_attr_mask); 2379 2380 /** 2381 * ib_query_qp - Returns the attribute list and current values for the 2382 * specified QP. 2383 * @qp: The QP to query. 2384 * @qp_attr: The attributes of the specified QP. 2385 * @qp_attr_mask: A bit-mask used to select specific attributes to query. 2386 * @qp_init_attr: Additional attributes of the selected QP. 2387 * 2388 * The qp_attr_mask may be used to limit the query to gathering only the 2389 * selected attributes. 2390 */ 2391 int ib_query_qp(struct ib_qp *qp, 2392 struct ib_qp_attr *qp_attr, 2393 int qp_attr_mask, 2394 struct ib_qp_init_attr *qp_init_attr); 2395 2396 /** 2397 * ib_destroy_qp - Destroys the specified QP. 2398 * @qp: The QP to destroy. 2399 */ 2400 int ib_destroy_qp(struct ib_qp *qp); 2401 2402 /** 2403 * ib_open_qp - Obtain a reference to an existing sharable QP. 2404 * @xrcd - XRC domain 2405 * @qp_open_attr: Attributes identifying the QP to open. 2406 * 2407 * Returns a reference to a sharable QP. 2408 */ 2409 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd, 2410 struct ib_qp_open_attr *qp_open_attr); 2411 2412 /** 2413 * ib_close_qp - Release an external reference to a QP. 2414 * @qp: The QP handle to release 2415 * 2416 * The opened QP handle is released by the caller. The underlying 2417 * shared QP is not destroyed until all internal references are released. 2418 */ 2419 int ib_close_qp(struct ib_qp *qp); 2420 2421 /** 2422 * ib_post_send - Posts a list of work requests to the send queue of 2423 * the specified QP. 2424 * @qp: The QP to post the work request on. 2425 * @send_wr: A list of work requests to post on the send queue. 2426 * @bad_send_wr: On an immediate failure, this parameter will reference 2427 * the work request that failed to be posted on the QP. 2428 * 2429 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate 2430 * error is returned, the QP state shall not be affected, 2431 * ib_post_send() will return an immediate error after queueing any 2432 * earlier work requests in the list. 2433 */ 2434 static inline int ib_post_send(struct ib_qp *qp, 2435 struct ib_send_wr *send_wr, 2436 struct ib_send_wr **bad_send_wr) 2437 { 2438 return qp->device->post_send(qp, send_wr, bad_send_wr); 2439 } 2440 2441 /** 2442 * ib_post_recv - Posts a list of work requests to the receive queue of 2443 * the specified QP. 2444 * @qp: The QP to post the work request on. 2445 * @recv_wr: A list of work requests to post on the receive queue. 2446 * @bad_recv_wr: On an immediate failure, this parameter will reference 2447 * the work request that failed to be posted on the QP. 2448 */ 2449 static inline int ib_post_recv(struct ib_qp *qp, 2450 struct ib_recv_wr *recv_wr, 2451 struct ib_recv_wr **bad_recv_wr) 2452 { 2453 return qp->device->post_recv(qp, recv_wr, bad_recv_wr); 2454 } 2455 2456 /** 2457 * ib_create_cq - Creates a CQ on the specified device. 2458 * @device: The device on which to create the CQ. 2459 * @comp_handler: A user-specified callback that is invoked when a 2460 * completion event occurs on the CQ. 2461 * @event_handler: A user-specified callback that is invoked when an 2462 * asynchronous event not associated with a completion occurs on the CQ. 2463 * @cq_context: Context associated with the CQ returned to the user via 2464 * the associated completion and event handlers. 2465 * @cq_attr: The attributes the CQ should be created upon. 2466 * 2467 * Users can examine the cq structure to determine the actual CQ size. 2468 */ 2469 struct ib_cq *ib_create_cq(struct ib_device *device, 2470 ib_comp_handler comp_handler, 2471 void (*event_handler)(struct ib_event *, void *), 2472 void *cq_context, 2473 const struct ib_cq_init_attr *cq_attr); 2474 2475 /** 2476 * ib_resize_cq - Modifies the capacity of the CQ. 2477 * @cq: The CQ to resize. 2478 * @cqe: The minimum size of the CQ. 2479 * 2480 * Users can examine the cq structure to determine the actual CQ size. 2481 */ 2482 int ib_resize_cq(struct ib_cq *cq, int cqe); 2483 2484 /** 2485 * ib_modify_cq - Modifies moderation params of the CQ 2486 * @cq: The CQ to modify. 2487 * @cq_count: number of CQEs that will trigger an event 2488 * @cq_period: max period of time in usec before triggering an event 2489 * 2490 */ 2491 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period); 2492 2493 /** 2494 * ib_destroy_cq - Destroys the specified CQ. 2495 * @cq: The CQ to destroy. 2496 */ 2497 int ib_destroy_cq(struct ib_cq *cq); 2498 2499 /** 2500 * ib_poll_cq - poll a CQ for completion(s) 2501 * @cq:the CQ being polled 2502 * @num_entries:maximum number of completions to return 2503 * @wc:array of at least @num_entries &struct ib_wc where completions 2504 * will be returned 2505 * 2506 * Poll a CQ for (possibly multiple) completions. If the return value 2507 * is < 0, an error occurred. If the return value is >= 0, it is the 2508 * number of completions returned. If the return value is 2509 * non-negative and < num_entries, then the CQ was emptied. 2510 */ 2511 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries, 2512 struct ib_wc *wc) 2513 { 2514 return cq->device->poll_cq(cq, num_entries, wc); 2515 } 2516 2517 /** 2518 * ib_peek_cq - Returns the number of unreaped completions currently 2519 * on the specified CQ. 2520 * @cq: The CQ to peek. 2521 * @wc_cnt: A minimum number of unreaped completions to check for. 2522 * 2523 * If the number of unreaped completions is greater than or equal to wc_cnt, 2524 * this function returns wc_cnt, otherwise, it returns the actual number of 2525 * unreaped completions. 2526 */ 2527 int ib_peek_cq(struct ib_cq *cq, int wc_cnt); 2528 2529 /** 2530 * ib_req_notify_cq - Request completion notification on a CQ. 2531 * @cq: The CQ to generate an event for. 2532 * @flags: 2533 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP 2534 * to request an event on the next solicited event or next work 2535 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS 2536 * may also be |ed in to request a hint about missed events, as 2537 * described below. 2538 * 2539 * Return Value: 2540 * < 0 means an error occurred while requesting notification 2541 * == 0 means notification was requested successfully, and if 2542 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events 2543 * were missed and it is safe to wait for another event. In 2544 * this case is it guaranteed that any work completions added 2545 * to the CQ since the last CQ poll will trigger a completion 2546 * notification event. 2547 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed 2548 * in. It means that the consumer must poll the CQ again to 2549 * make sure it is empty to avoid missing an event because of a 2550 * race between requesting notification and an entry being 2551 * added to the CQ. This return value means it is possible 2552 * (but not guaranteed) that a work completion has been added 2553 * to the CQ since the last poll without triggering a 2554 * completion notification event. 2555 */ 2556 static inline int ib_req_notify_cq(struct ib_cq *cq, 2557 enum ib_cq_notify_flags flags) 2558 { 2559 return cq->device->req_notify_cq(cq, flags); 2560 } 2561 2562 /** 2563 * ib_req_ncomp_notif - Request completion notification when there are 2564 * at least the specified number of unreaped completions on the CQ. 2565 * @cq: The CQ to generate an event for. 2566 * @wc_cnt: The number of unreaped completions that should be on the 2567 * CQ before an event is generated. 2568 */ 2569 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt) 2570 { 2571 return cq->device->req_ncomp_notif ? 2572 cq->device->req_ncomp_notif(cq, wc_cnt) : 2573 -ENOSYS; 2574 } 2575 2576 /** 2577 * ib_get_dma_mr - Returns a memory region for system memory that is 2578 * usable for DMA. 2579 * @pd: The protection domain associated with the memory region. 2580 * @mr_access_flags: Specifies the memory access rights. 2581 * 2582 * Note that the ib_dma_*() functions defined below must be used 2583 * to create/destroy addresses used with the Lkey or Rkey returned 2584 * by ib_get_dma_mr(). 2585 */ 2586 struct ib_mr *ib_get_dma_mr(struct ib_pd *pd, int mr_access_flags); 2587 2588 /** 2589 * ib_dma_mapping_error - check a DMA addr for error 2590 * @dev: The device for which the dma_addr was created 2591 * @dma_addr: The DMA address to check 2592 */ 2593 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr) 2594 { 2595 if (dev->dma_ops) 2596 return dev->dma_ops->mapping_error(dev, dma_addr); 2597 return dma_mapping_error(dev->dma_device, dma_addr); 2598 } 2599 2600 /** 2601 * ib_dma_map_single - Map a kernel virtual address to DMA address 2602 * @dev: The device for which the dma_addr is to be created 2603 * @cpu_addr: The kernel virtual address 2604 * @size: The size of the region in bytes 2605 * @direction: The direction of the DMA 2606 */ 2607 static inline u64 ib_dma_map_single(struct ib_device *dev, 2608 void *cpu_addr, size_t size, 2609 enum dma_data_direction direction) 2610 { 2611 if (dev->dma_ops) 2612 return dev->dma_ops->map_single(dev, cpu_addr, size, direction); 2613 return dma_map_single(dev->dma_device, cpu_addr, size, direction); 2614 } 2615 2616 /** 2617 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single() 2618 * @dev: The device for which the DMA address was created 2619 * @addr: The DMA address 2620 * @size: The size of the region in bytes 2621 * @direction: The direction of the DMA 2622 */ 2623 static inline void ib_dma_unmap_single(struct ib_device *dev, 2624 u64 addr, size_t size, 2625 enum dma_data_direction direction) 2626 { 2627 if (dev->dma_ops) 2628 dev->dma_ops->unmap_single(dev, addr, size, direction); 2629 else 2630 dma_unmap_single(dev->dma_device, addr, size, direction); 2631 } 2632 2633 static inline u64 ib_dma_map_single_attrs(struct ib_device *dev, 2634 void *cpu_addr, size_t size, 2635 enum dma_data_direction direction, 2636 struct dma_attrs *attrs) 2637 { 2638 return dma_map_single_attrs(dev->dma_device, cpu_addr, size, 2639 direction, attrs); 2640 } 2641 2642 static inline void ib_dma_unmap_single_attrs(struct ib_device *dev, 2643 u64 addr, size_t size, 2644 enum dma_data_direction direction, 2645 struct dma_attrs *attrs) 2646 { 2647 return dma_unmap_single_attrs(dev->dma_device, addr, size, 2648 direction, attrs); 2649 } 2650 2651 /** 2652 * ib_dma_map_page - Map a physical page to DMA address 2653 * @dev: The device for which the dma_addr is to be created 2654 * @page: The page to be mapped 2655 * @offset: The offset within the page 2656 * @size: The size of the region in bytes 2657 * @direction: The direction of the DMA 2658 */ 2659 static inline u64 ib_dma_map_page(struct ib_device *dev, 2660 struct page *page, 2661 unsigned long offset, 2662 size_t size, 2663 enum dma_data_direction direction) 2664 { 2665 if (dev->dma_ops) 2666 return dev->dma_ops->map_page(dev, page, offset, size, direction); 2667 return dma_map_page(dev->dma_device, page, offset, size, direction); 2668 } 2669 2670 /** 2671 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page() 2672 * @dev: The device for which the DMA address was created 2673 * @addr: The DMA address 2674 * @size: The size of the region in bytes 2675 * @direction: The direction of the DMA 2676 */ 2677 static inline void ib_dma_unmap_page(struct ib_device *dev, 2678 u64 addr, size_t size, 2679 enum dma_data_direction direction) 2680 { 2681 if (dev->dma_ops) 2682 dev->dma_ops->unmap_page(dev, addr, size, direction); 2683 else 2684 dma_unmap_page(dev->dma_device, addr, size, direction); 2685 } 2686 2687 /** 2688 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses 2689 * @dev: The device for which the DMA addresses are to be created 2690 * @sg: The array of scatter/gather entries 2691 * @nents: The number of scatter/gather entries 2692 * @direction: The direction of the DMA 2693 */ 2694 static inline int ib_dma_map_sg(struct ib_device *dev, 2695 struct scatterlist *sg, int nents, 2696 enum dma_data_direction direction) 2697 { 2698 if (dev->dma_ops) 2699 return dev->dma_ops->map_sg(dev, sg, nents, direction); 2700 return dma_map_sg(dev->dma_device, sg, nents, direction); 2701 } 2702 2703 /** 2704 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses 2705 * @dev: The device for which the DMA addresses were created 2706 * @sg: The array of scatter/gather entries 2707 * @nents: The number of scatter/gather entries 2708 * @direction: The direction of the DMA 2709 */ 2710 static inline void ib_dma_unmap_sg(struct ib_device *dev, 2711 struct scatterlist *sg, int nents, 2712 enum dma_data_direction direction) 2713 { 2714 if (dev->dma_ops) 2715 dev->dma_ops->unmap_sg(dev, sg, nents, direction); 2716 else 2717 dma_unmap_sg(dev->dma_device, sg, nents, direction); 2718 } 2719 2720 static inline int ib_dma_map_sg_attrs(struct ib_device *dev, 2721 struct scatterlist *sg, int nents, 2722 enum dma_data_direction direction, 2723 struct dma_attrs *attrs) 2724 { 2725 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, attrs); 2726 } 2727 2728 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev, 2729 struct scatterlist *sg, int nents, 2730 enum dma_data_direction direction, 2731 struct dma_attrs *attrs) 2732 { 2733 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, attrs); 2734 } 2735 /** 2736 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry 2737 * @dev: The device for which the DMA addresses were created 2738 * @sg: The scatter/gather entry 2739 * 2740 * Note: this function is obsolete. To do: change all occurrences of 2741 * ib_sg_dma_address() into sg_dma_address(). 2742 */ 2743 static inline u64 ib_sg_dma_address(struct ib_device *dev, 2744 struct scatterlist *sg) 2745 { 2746 return sg_dma_address(sg); 2747 } 2748 2749 /** 2750 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry 2751 * @dev: The device for which the DMA addresses were created 2752 * @sg: The scatter/gather entry 2753 * 2754 * Note: this function is obsolete. To do: change all occurrences of 2755 * ib_sg_dma_len() into sg_dma_len(). 2756 */ 2757 static inline unsigned int ib_sg_dma_len(struct ib_device *dev, 2758 struct scatterlist *sg) 2759 { 2760 return sg_dma_len(sg); 2761 } 2762 2763 /** 2764 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU 2765 * @dev: The device for which the DMA address was created 2766 * @addr: The DMA address 2767 * @size: The size of the region in bytes 2768 * @dir: The direction of the DMA 2769 */ 2770 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev, 2771 u64 addr, 2772 size_t size, 2773 enum dma_data_direction dir) 2774 { 2775 if (dev->dma_ops) 2776 dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir); 2777 else 2778 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir); 2779 } 2780 2781 /** 2782 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device 2783 * @dev: The device for which the DMA address was created 2784 * @addr: The DMA address 2785 * @size: The size of the region in bytes 2786 * @dir: The direction of the DMA 2787 */ 2788 static inline void ib_dma_sync_single_for_device(struct ib_device *dev, 2789 u64 addr, 2790 size_t size, 2791 enum dma_data_direction dir) 2792 { 2793 if (dev->dma_ops) 2794 dev->dma_ops->sync_single_for_device(dev, addr, size, dir); 2795 else 2796 dma_sync_single_for_device(dev->dma_device, addr, size, dir); 2797 } 2798 2799 /** 2800 * ib_dma_alloc_coherent - Allocate memory and map it for DMA 2801 * @dev: The device for which the DMA address is requested 2802 * @size: The size of the region to allocate in bytes 2803 * @dma_handle: A pointer for returning the DMA address of the region 2804 * @flag: memory allocator flags 2805 */ 2806 static inline void *ib_dma_alloc_coherent(struct ib_device *dev, 2807 size_t size, 2808 u64 *dma_handle, 2809 gfp_t flag) 2810 { 2811 if (dev->dma_ops) 2812 return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag); 2813 else { 2814 dma_addr_t handle; 2815 void *ret; 2816 2817 ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag); 2818 *dma_handle = handle; 2819 return ret; 2820 } 2821 } 2822 2823 /** 2824 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent() 2825 * @dev: The device for which the DMA addresses were allocated 2826 * @size: The size of the region 2827 * @cpu_addr: the address returned by ib_dma_alloc_coherent() 2828 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent() 2829 */ 2830 static inline void ib_dma_free_coherent(struct ib_device *dev, 2831 size_t size, void *cpu_addr, 2832 u64 dma_handle) 2833 { 2834 if (dev->dma_ops) 2835 dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle); 2836 else 2837 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle); 2838 } 2839 2840 /** 2841 * ib_query_mr - Retrieves information about a specific memory region. 2842 * @mr: The memory region to retrieve information about. 2843 * @mr_attr: The attributes of the specified memory region. 2844 */ 2845 int ib_query_mr(struct ib_mr *mr, struct ib_mr_attr *mr_attr); 2846 2847 /** 2848 * ib_dereg_mr - Deregisters a memory region and removes it from the 2849 * HCA translation table. 2850 * @mr: The memory region to deregister. 2851 * 2852 * This function can fail, if the memory region has memory windows bound to it. 2853 */ 2854 int ib_dereg_mr(struct ib_mr *mr); 2855 2856 struct ib_mr *ib_alloc_mr(struct ib_pd *pd, 2857 enum ib_mr_type mr_type, 2858 u32 max_num_sg); 2859 2860 /** 2861 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR 2862 * R_Key and L_Key. 2863 * @mr - struct ib_mr pointer to be updated. 2864 * @newkey - new key to be used. 2865 */ 2866 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey) 2867 { 2868 mr->lkey = (mr->lkey & 0xffffff00) | newkey; 2869 mr->rkey = (mr->rkey & 0xffffff00) | newkey; 2870 } 2871 2872 /** 2873 * ib_inc_rkey - increments the key portion of the given rkey. Can be used 2874 * for calculating a new rkey for type 2 memory windows. 2875 * @rkey - the rkey to increment. 2876 */ 2877 static inline u32 ib_inc_rkey(u32 rkey) 2878 { 2879 const u32 mask = 0x000000ff; 2880 return ((rkey + 1) & mask) | (rkey & ~mask); 2881 } 2882 2883 /** 2884 * ib_alloc_mw - Allocates a memory window. 2885 * @pd: The protection domain associated with the memory window. 2886 * @type: The type of the memory window (1 or 2). 2887 */ 2888 struct ib_mw *ib_alloc_mw(struct ib_pd *pd, enum ib_mw_type type); 2889 2890 /** 2891 * ib_bind_mw - Posts a work request to the send queue of the specified 2892 * QP, which binds the memory window to the given address range and 2893 * remote access attributes. 2894 * @qp: QP to post the bind work request on. 2895 * @mw: The memory window to bind. 2896 * @mw_bind: Specifies information about the memory window, including 2897 * its address range, remote access rights, and associated memory region. 2898 * 2899 * If there is no immediate error, the function will update the rkey member 2900 * of the mw parameter to its new value. The bind operation can still fail 2901 * asynchronously. 2902 */ 2903 static inline int ib_bind_mw(struct ib_qp *qp, 2904 struct ib_mw *mw, 2905 struct ib_mw_bind *mw_bind) 2906 { 2907 /* XXX reference counting in corresponding MR? */ 2908 return mw->device->bind_mw ? 2909 mw->device->bind_mw(qp, mw, mw_bind) : 2910 -ENOSYS; 2911 } 2912 2913 /** 2914 * ib_dealloc_mw - Deallocates a memory window. 2915 * @mw: The memory window to deallocate. 2916 */ 2917 int ib_dealloc_mw(struct ib_mw *mw); 2918 2919 /** 2920 * ib_alloc_fmr - Allocates a unmapped fast memory region. 2921 * @pd: The protection domain associated with the unmapped region. 2922 * @mr_access_flags: Specifies the memory access rights. 2923 * @fmr_attr: Attributes of the unmapped region. 2924 * 2925 * A fast memory region must be mapped before it can be used as part of 2926 * a work request. 2927 */ 2928 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd, 2929 int mr_access_flags, 2930 struct ib_fmr_attr *fmr_attr); 2931 2932 /** 2933 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region. 2934 * @fmr: The fast memory region to associate with the pages. 2935 * @page_list: An array of physical pages to map to the fast memory region. 2936 * @list_len: The number of pages in page_list. 2937 * @iova: The I/O virtual address to use with the mapped region. 2938 */ 2939 static inline int ib_map_phys_fmr(struct ib_fmr *fmr, 2940 u64 *page_list, int list_len, 2941 u64 iova) 2942 { 2943 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova); 2944 } 2945 2946 /** 2947 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions. 2948 * @fmr_list: A linked list of fast memory regions to unmap. 2949 */ 2950 int ib_unmap_fmr(struct list_head *fmr_list); 2951 2952 /** 2953 * ib_dealloc_fmr - Deallocates a fast memory region. 2954 * @fmr: The fast memory region to deallocate. 2955 */ 2956 int ib_dealloc_fmr(struct ib_fmr *fmr); 2957 2958 /** 2959 * ib_attach_mcast - Attaches the specified QP to a multicast group. 2960 * @qp: QP to attach to the multicast group. The QP must be type 2961 * IB_QPT_UD. 2962 * @gid: Multicast group GID. 2963 * @lid: Multicast group LID in host byte order. 2964 * 2965 * In order to send and receive multicast packets, subnet 2966 * administration must have created the multicast group and configured 2967 * the fabric appropriately. The port associated with the specified 2968 * QP must also be a member of the multicast group. 2969 */ 2970 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid); 2971 2972 /** 2973 * ib_detach_mcast - Detaches the specified QP from a multicast group. 2974 * @qp: QP to detach from the multicast group. 2975 * @gid: Multicast group GID. 2976 * @lid: Multicast group LID in host byte order. 2977 */ 2978 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid); 2979 2980 /** 2981 * ib_alloc_xrcd - Allocates an XRC domain. 2982 * @device: The device on which to allocate the XRC domain. 2983 */ 2984 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device); 2985 2986 /** 2987 * ib_dealloc_xrcd - Deallocates an XRC domain. 2988 * @xrcd: The XRC domain to deallocate. 2989 */ 2990 int ib_dealloc_xrcd(struct ib_xrcd *xrcd); 2991 2992 struct ib_flow *ib_create_flow(struct ib_qp *qp, 2993 struct ib_flow_attr *flow_attr, int domain); 2994 int ib_destroy_flow(struct ib_flow *flow_id); 2995 2996 static inline int ib_check_mr_access(int flags) 2997 { 2998 /* 2999 * Local write permission is required if remote write or 3000 * remote atomic permission is also requested. 3001 */ 3002 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) && 3003 !(flags & IB_ACCESS_LOCAL_WRITE)) 3004 return -EINVAL; 3005 3006 return 0; 3007 } 3008 3009 /** 3010 * ib_check_mr_status: lightweight check of MR status. 3011 * This routine may provide status checks on a selected 3012 * ib_mr. first use is for signature status check. 3013 * 3014 * @mr: A memory region. 3015 * @check_mask: Bitmask of which checks to perform from 3016 * ib_mr_status_check enumeration. 3017 * @mr_status: The container of relevant status checks. 3018 * failed checks will be indicated in the status bitmask 3019 * and the relevant info shall be in the error item. 3020 */ 3021 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask, 3022 struct ib_mr_status *mr_status); 3023 3024 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port, 3025 u16 pkey, const union ib_gid *gid, 3026 const struct sockaddr *addr); 3027 3028 int ib_map_mr_sg(struct ib_mr *mr, 3029 struct scatterlist *sg, 3030 int sg_nents, 3031 unsigned int page_size); 3032 3033 static inline int 3034 ib_map_mr_sg_zbva(struct ib_mr *mr, 3035 struct scatterlist *sg, 3036 int sg_nents, 3037 unsigned int page_size) 3038 { 3039 int n; 3040 3041 n = ib_map_mr_sg(mr, sg, sg_nents, page_size); 3042 mr->iova = 0; 3043 3044 return n; 3045 } 3046 3047 int ib_sg_to_pages(struct ib_mr *mr, 3048 struct scatterlist *sgl, 3049 int sg_nents, 3050 int (*set_page)(struct ib_mr *, u64)); 3051 3052 #endif /* IB_VERBS_H */ 3053