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 * $Id: ib_verbs.h 1349 2004-12-16 21:09:43Z roland $ 39 * 40 * $FreeBSD$ 41 */ 42 43 44#if !defined(IB_VERBS_H) 45#define IB_VERBS_H 46 47#include <contrib/rdma/types.h> 48#include <sys/lock.h> 49#include <sys/mutex.h> 50 51struct rdma_scatterlist { 52 void *page; 53 unsigned int length; 54 unsigned int offset; 55}; 56struct vm_object; 57 58union ib_gid { 59 u8 raw[16]; 60 struct { 61 __be64 subnet_prefix; 62 __be64 interface_id; 63 } global; 64}; 65 66enum rdma_node_type { 67 /* IB values map to NodeInfo:NodeType. */ 68 RDMA_NODE_IB_CA = 1, 69 RDMA_NODE_IB_SWITCH, 70 RDMA_NODE_IB_ROUTER, 71 RDMA_NODE_RNIC 72}; 73 74enum rdma_transport_type { 75 RDMA_TRANSPORT_IB, 76 RDMA_TRANSPORT_IWARP 77}; 78 79enum rdma_transport_type 80rdma_node_get_transport(enum rdma_node_type node_type); 81 82enum ib_device_cap_flags { 83 IB_DEVICE_RESIZE_MAX_WR = 1, 84 IB_DEVICE_BAD_PKEY_CNTR = (1<<1), 85 IB_DEVICE_BAD_QKEY_CNTR = (1<<2), 86 IB_DEVICE_RAW_MULTI = (1<<3), 87 IB_DEVICE_AUTO_PATH_MIG = (1<<4), 88 IB_DEVICE_CHANGE_PHY_PORT = (1<<5), 89 IB_DEVICE_UD_AV_PORT_ENFORCE = (1<<6), 90 IB_DEVICE_CURR_QP_STATE_MOD = (1<<7), 91 IB_DEVICE_SHUTDOWN_PORT = (1<<8), 92 IB_DEVICE_INIT_TYPE = (1<<9), 93 IB_DEVICE_PORT_ACTIVE_EVENT = (1<<10), 94 IB_DEVICE_SYS_IMAGE_GUID = (1<<11), 95 IB_DEVICE_RC_RNR_NAK_GEN = (1<<12), 96 IB_DEVICE_SRQ_RESIZE = (1<<13), 97 IB_DEVICE_N_NOTIFY_CQ = (1<<14), 98 IB_DEVICE_ZERO_STAG = (1<<15), 99 IB_DEVICE_SEND_W_INV = (1<<16), 100 IB_DEVICE_MEM_WINDOW = (1<<17) 101}; 102 103enum ib_atomic_cap { 104 IB_ATOMIC_NONE, 105 IB_ATOMIC_HCA, 106 IB_ATOMIC_GLOB 107}; 108 109struct ib_device_attr { 110 u64 fw_ver; 111 __be64 sys_image_guid; 112 u64 max_mr_size; 113 u64 page_size_cap; 114 u32 vendor_id; 115 u32 vendor_part_id; 116 u32 hw_ver; 117 int max_qp; 118 int max_qp_wr; 119 int device_cap_flags; 120 int max_sge; 121 int max_sge_rd; 122 int max_cq; 123 int max_cqe; 124 int max_mr; 125 int max_pd; 126 int max_qp_rd_atom; 127 int max_ee_rd_atom; 128 int max_res_rd_atom; 129 int max_qp_init_rd_atom; 130 int max_ee_init_rd_atom; 131 enum ib_atomic_cap atomic_cap; 132 int max_ee; 133 int max_rdd; 134 int max_mw; 135 int max_raw_ipv6_qp; 136 int max_raw_ethy_qp; 137 int max_mcast_grp; 138 int max_mcast_qp_attach; 139 int max_total_mcast_qp_attach; 140 int max_ah; 141 int max_fmr; 142 int max_map_per_fmr; 143 int max_srq; 144 int max_srq_wr; 145 int max_srq_sge; 146 u16 max_pkeys; 147 u8 local_ca_ack_delay; 148}; 149 150enum ib_mtu { 151 IB_MTU_256 = 1, 152 IB_MTU_512 = 2, 153 IB_MTU_1024 = 3, 154 IB_MTU_2048 = 4, 155 IB_MTU_4096 = 5 156}; 157 158static inline int ib_mtu_enum_to_int(enum ib_mtu mtu) 159{ 160 switch (mtu) { 161 case IB_MTU_256: return 256; 162 case IB_MTU_512: return 512; 163 case IB_MTU_1024: return 1024; 164 case IB_MTU_2048: return 2048; 165 case IB_MTU_4096: return 4096; 166 default: return -1; 167 } 168} 169 170enum ib_port_state { 171 IB_PORT_NOP = 0, 172 IB_PORT_DOWN = 1, 173 IB_PORT_INIT = 2, 174 IB_PORT_ARMED = 3, 175 IB_PORT_ACTIVE = 4, 176 IB_PORT_ACTIVE_DEFER = 5 177}; 178 179enum ib_port_cap_flags { 180 IB_PORT_SM = 1 << 1, 181 IB_PORT_NOTICE_SUP = 1 << 2, 182 IB_PORT_TRAP_SUP = 1 << 3, 183 IB_PORT_OPT_IPD_SUP = 1 << 4, 184 IB_PORT_AUTO_MIGR_SUP = 1 << 5, 185 IB_PORT_SL_MAP_SUP = 1 << 6, 186 IB_PORT_MKEY_NVRAM = 1 << 7, 187 IB_PORT_PKEY_NVRAM = 1 << 8, 188 IB_PORT_LED_INFO_SUP = 1 << 9, 189 IB_PORT_SM_DISABLED = 1 << 10, 190 IB_PORT_SYS_IMAGE_GUID_SUP = 1 << 11, 191 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP = 1 << 12, 192 IB_PORT_CM_SUP = 1 << 16, 193 IB_PORT_SNMP_TUNNEL_SUP = 1 << 17, 194 IB_PORT_REINIT_SUP = 1 << 18, 195 IB_PORT_DEVICE_MGMT_SUP = 1 << 19, 196 IB_PORT_VENDOR_CLASS_SUP = 1 << 20, 197 IB_PORT_DR_NOTICE_SUP = 1 << 21, 198 IB_PORT_CAP_MASK_NOTICE_SUP = 1 << 22, 199 IB_PORT_BOOT_MGMT_SUP = 1 << 23, 200 IB_PORT_LINK_LATENCY_SUP = 1 << 24, 201 IB_PORT_CLIENT_REG_SUP = 1 << 25 202}; 203 204enum ib_port_width { 205 IB_WIDTH_1X = 1, 206 IB_WIDTH_4X = 2, 207 IB_WIDTH_8X = 4, 208 IB_WIDTH_12X = 8 209}; 210 211static inline int ib_width_enum_to_int(enum ib_port_width width) 212{ 213 switch (width) { 214 case IB_WIDTH_1X: return 1; 215 case IB_WIDTH_4X: return 4; 216 case IB_WIDTH_8X: return 8; 217 case IB_WIDTH_12X: return 12; 218 default: return -1; 219 } 220} 221 222struct ib_port_attr { 223 enum ib_port_state state; 224 enum ib_mtu max_mtu; 225 enum ib_mtu active_mtu; 226 int gid_tbl_len; 227 u32 port_cap_flags; 228 u32 max_msg_sz; 229 u32 bad_pkey_cntr; 230 u32 qkey_viol_cntr; 231 u16 pkey_tbl_len; 232 u16 lid; 233 u16 sm_lid; 234 u8 lmc; 235 u8 max_vl_num; 236 u8 sm_sl; 237 u8 subnet_timeout; 238 u8 init_type_reply; 239 u8 active_width; 240 u8 active_speed; 241 u8 phys_state; 242}; 243 244enum ib_device_modify_flags { 245 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0, 246 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1 247}; 248 249struct ib_device_modify { 250 u64 sys_image_guid; 251 char node_desc[64]; 252}; 253 254enum ib_port_modify_flags { 255 IB_PORT_SHUTDOWN = 1, 256 IB_PORT_INIT_TYPE = (1<<2), 257 IB_PORT_RESET_QKEY_CNTR = (1<<3) 258}; 259 260struct ib_port_modify { 261 u32 set_port_cap_mask; 262 u32 clr_port_cap_mask; 263 u8 init_type; 264}; 265 266enum ib_event_type { 267 IB_EVENT_CQ_ERR, 268 IB_EVENT_QP_FATAL, 269 IB_EVENT_QP_REQ_ERR, 270 IB_EVENT_QP_ACCESS_ERR, 271 IB_EVENT_COMM_EST, 272 IB_EVENT_SQ_DRAINED, 273 IB_EVENT_PATH_MIG, 274 IB_EVENT_PATH_MIG_ERR, 275 IB_EVENT_DEVICE_FATAL, 276 IB_EVENT_PORT_ACTIVE, 277 IB_EVENT_PORT_ERR, 278 IB_EVENT_LID_CHANGE, 279 IB_EVENT_PKEY_CHANGE, 280 IB_EVENT_SM_CHANGE, 281 IB_EVENT_SRQ_ERR, 282 IB_EVENT_SRQ_LIMIT_REACHED, 283 IB_EVENT_QP_LAST_WQE_REACHED, 284 IB_EVENT_CLIENT_REREGISTER 285}; 286 287enum dma_data_direction { 288 DMA_BIDIRECTIONAL = 0, 289 DMA_TO_DEVICE = 1, 290 DMA_FROM_DEVICE = 2, 291 DMA_NONE = 3, 292}; 293 294struct ib_event { 295 struct ib_device *device; 296 union { 297 struct ib_cq *cq; 298 struct ib_qp *qp; 299 struct ib_srq *srq; 300 u8 port_num; 301 } element; 302 enum ib_event_type event; 303}; 304 305struct ib_event_handler { 306 struct ib_device *device; 307 void (*handler)(struct ib_event_handler *, struct ib_event *); 308 TAILQ_ENTRY(ib_event_handler) list; 309}; 310 311#define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \ 312 do { \ 313 (_ptr)->device = _device; \ 314 (_ptr)->handler = _handler; \ 315 } while (0) 316 317struct ib_global_route { 318 union ib_gid dgid; 319 u32 flow_label; 320 u8 sgid_index; 321 u8 hop_limit; 322 u8 traffic_class; 323}; 324 325struct ib_grh { 326 __be32 version_tclass_flow; 327 __be16 paylen; 328 u8 next_hdr; 329 u8 hop_limit; 330 union ib_gid sgid; 331 union ib_gid dgid; 332}; 333 334enum { 335 IB_MULTICAST_QPN = 0xffffff 336}; 337 338#define IB_LID_PERMISSIVE __constant_htons(0xFFFF) 339 340enum ib_ah_flags { 341 IB_AH_GRH = 1 342}; 343 344enum ib_rate { 345 IB_RATE_PORT_CURRENT = 0, 346 IB_RATE_2_5_GBPS = 2, 347 IB_RATE_5_GBPS = 5, 348 IB_RATE_10_GBPS = 3, 349 IB_RATE_20_GBPS = 6, 350 IB_RATE_30_GBPS = 4, 351 IB_RATE_40_GBPS = 7, 352 IB_RATE_60_GBPS = 8, 353 IB_RATE_80_GBPS = 9, 354 IB_RATE_120_GBPS = 10 355}; 356 357/** 358 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the 359 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be 360 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec. 361 * @rate: rate to convert. 362 */ 363int ib_rate_to_mult(enum ib_rate rate); 364 365/** 366 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate 367 * enum. 368 * @mult: multiple to convert. 369 */ 370enum ib_rate mult_to_ib_rate(int mult); 371 372struct ib_ah_attr { 373 struct ib_global_route grh; 374 u16 dlid; 375 u8 sl; 376 u8 src_path_bits; 377 u8 static_rate; 378 u8 ah_flags; 379 u8 port_num; 380}; 381 382enum ib_wc_status { 383 IB_WC_SUCCESS, 384 IB_WC_LOC_LEN_ERR, 385 IB_WC_LOC_QP_OP_ERR, 386 IB_WC_LOC_EEC_OP_ERR, 387 IB_WC_LOC_PROT_ERR, 388 IB_WC_WR_FLUSH_ERR, 389 IB_WC_MW_BIND_ERR, 390 IB_WC_BAD_RESP_ERR, 391 IB_WC_LOC_ACCESS_ERR, 392 IB_WC_REM_INV_REQ_ERR, 393 IB_WC_REM_ACCESS_ERR, 394 IB_WC_REM_OP_ERR, 395 IB_WC_RETRY_EXC_ERR, 396 IB_WC_RNR_RETRY_EXC_ERR, 397 IB_WC_LOC_RDD_VIOL_ERR, 398 IB_WC_REM_INV_RD_REQ_ERR, 399 IB_WC_REM_ABORT_ERR, 400 IB_WC_INV_EECN_ERR, 401 IB_WC_INV_EEC_STATE_ERR, 402 IB_WC_FATAL_ERR, 403 IB_WC_RESP_TIMEOUT_ERR, 404 IB_WC_GENERAL_ERR 405}; 406 407enum ib_wc_opcode { 408 IB_WC_SEND, 409 IB_WC_RDMA_WRITE, 410 IB_WC_RDMA_READ, 411 IB_WC_COMP_SWAP, 412 IB_WC_FETCH_ADD, 413 IB_WC_BIND_MW, 414/* 415 * Set value of IB_WC_RECV so consumers can test if a completion is a 416 * receive by testing (opcode & IB_WC_RECV). 417 */ 418 IB_WC_RECV = 1 << 7, 419 IB_WC_RECV_RDMA_WITH_IMM 420}; 421 422enum ib_wc_flags { 423 IB_WC_GRH = 1, 424 IB_WC_WITH_IMM = (1<<1) 425}; 426 427struct ib_wc { 428 u64 wr_id; 429 enum ib_wc_status status; 430 enum ib_wc_opcode opcode; 431 u32 vendor_err; 432 u32 byte_len; 433 struct ib_qp *qp; 434 __be32 imm_data; 435 u32 src_qp; 436 int wc_flags; 437 u16 pkey_index; 438 u16 slid; 439 u8 sl; 440 u8 dlid_path_bits; 441 u8 port_num; /* valid only for DR SMPs on switches */ 442}; 443 444enum ib_cq_notify_flags { 445 IB_CQ_SOLICITED = 1 << 0, 446 IB_CQ_NEXT_COMP = 1 << 1, 447 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP, 448 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2, 449}; 450 451enum ib_srq_attr_mask { 452 IB_SRQ_MAX_WR = 1 << 0, 453 IB_SRQ_LIMIT = 1 << 1, 454}; 455 456struct ib_srq_attr { 457 u32 max_wr; 458 u32 max_sge; 459 u32 srq_limit; 460}; 461 462struct ib_srq_init_attr { 463 void (*event_handler)(struct ib_event *, void *); 464 void *srq_context; 465 struct ib_srq_attr attr; 466}; 467 468struct ib_qp_cap { 469 u32 max_send_wr; 470 u32 max_recv_wr; 471 u32 max_send_sge; 472 u32 max_recv_sge; 473 u32 max_inline_data; 474}; 475 476enum ib_sig_type { 477 IB_SIGNAL_ALL_WR, 478 IB_SIGNAL_REQ_WR 479}; 480 481enum ib_qp_type { 482 /* 483 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries 484 * here (and in that order) since the MAD layer uses them as 485 * indices into a 2-entry table. 486 */ 487 IB_QPT_SMI, 488 IB_QPT_GSI, 489 490 IB_QPT_RC, 491 IB_QPT_UC, 492 IB_QPT_UD, 493 IB_QPT_RAW_IPV6, 494 IB_QPT_RAW_ETY 495}; 496 497struct ib_qp_init_attr { 498 void (*event_handler)(struct ib_event *, void *); 499 void *qp_context; 500 struct ib_cq *send_cq; 501 struct ib_cq *recv_cq; 502 struct ib_srq *srq; 503 struct ib_qp_cap cap; 504 enum ib_sig_type sq_sig_type; 505 enum ib_qp_type qp_type; 506 u8 port_num; /* special QP types only */ 507}; 508 509enum ib_rnr_timeout { 510 IB_RNR_TIMER_655_36 = 0, 511 IB_RNR_TIMER_000_01 = 1, 512 IB_RNR_TIMER_000_02 = 2, 513 IB_RNR_TIMER_000_03 = 3, 514 IB_RNR_TIMER_000_04 = 4, 515 IB_RNR_TIMER_000_06 = 5, 516 IB_RNR_TIMER_000_08 = 6, 517 IB_RNR_TIMER_000_12 = 7, 518 IB_RNR_TIMER_000_16 = 8, 519 IB_RNR_TIMER_000_24 = 9, 520 IB_RNR_TIMER_000_32 = 10, 521 IB_RNR_TIMER_000_48 = 11, 522 IB_RNR_TIMER_000_64 = 12, 523 IB_RNR_TIMER_000_96 = 13, 524 IB_RNR_TIMER_001_28 = 14, 525 IB_RNR_TIMER_001_92 = 15, 526 IB_RNR_TIMER_002_56 = 16, 527 IB_RNR_TIMER_003_84 = 17, 528 IB_RNR_TIMER_005_12 = 18, 529 IB_RNR_TIMER_007_68 = 19, 530 IB_RNR_TIMER_010_24 = 20, 531 IB_RNR_TIMER_015_36 = 21, 532 IB_RNR_TIMER_020_48 = 22, 533 IB_RNR_TIMER_030_72 = 23, 534 IB_RNR_TIMER_040_96 = 24, 535 IB_RNR_TIMER_061_44 = 25, 536 IB_RNR_TIMER_081_92 = 26, 537 IB_RNR_TIMER_122_88 = 27, 538 IB_RNR_TIMER_163_84 = 28, 539 IB_RNR_TIMER_245_76 = 29, 540 IB_RNR_TIMER_327_68 = 30, 541 IB_RNR_TIMER_491_52 = 31 542}; 543 544enum ib_qp_attr_mask { 545 IB_QP_STATE = 1, 546 IB_QP_CUR_STATE = (1<<1), 547 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2), 548 IB_QP_ACCESS_FLAGS = (1<<3), 549 IB_QP_PKEY_INDEX = (1<<4), 550 IB_QP_PORT = (1<<5), 551 IB_QP_QKEY = (1<<6), 552 IB_QP_AV = (1<<7), 553 IB_QP_PATH_MTU = (1<<8), 554 IB_QP_TIMEOUT = (1<<9), 555 IB_QP_RETRY_CNT = (1<<10), 556 IB_QP_RNR_RETRY = (1<<11), 557 IB_QP_RQ_PSN = (1<<12), 558 IB_QP_MAX_QP_RD_ATOMIC = (1<<13), 559 IB_QP_ALT_PATH = (1<<14), 560 IB_QP_MIN_RNR_TIMER = (1<<15), 561 IB_QP_SQ_PSN = (1<<16), 562 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17), 563 IB_QP_PATH_MIG_STATE = (1<<18), 564 IB_QP_CAP = (1<<19), 565 IB_QP_DEST_QPN = (1<<20) 566}; 567 568enum ib_qp_state { 569 IB_QPS_RESET, 570 IB_QPS_INIT, 571 IB_QPS_RTR, 572 IB_QPS_RTS, 573 IB_QPS_SQD, 574 IB_QPS_SQE, 575 IB_QPS_ERR 576}; 577 578enum ib_mig_state { 579 IB_MIG_MIGRATED, 580 IB_MIG_REARM, 581 IB_MIG_ARMED 582}; 583 584struct ib_qp_attr { 585 enum ib_qp_state qp_state; 586 enum ib_qp_state cur_qp_state; 587 enum ib_mtu path_mtu; 588 enum ib_mig_state path_mig_state; 589 u32 qkey; 590 u32 rq_psn; 591 u32 sq_psn; 592 u32 dest_qp_num; 593 int qp_access_flags; 594 struct ib_qp_cap cap; 595 struct ib_ah_attr ah_attr; 596 struct ib_ah_attr alt_ah_attr; 597 u16 pkey_index; 598 u16 alt_pkey_index; 599 u8 en_sqd_async_notify; 600 u8 sq_draining; 601 u8 max_rd_atomic; 602 u8 max_dest_rd_atomic; 603 u8 min_rnr_timer; 604 u8 port_num; 605 u8 timeout; 606 u8 retry_cnt; 607 u8 rnr_retry; 608 u8 alt_port_num; 609 u8 alt_timeout; 610}; 611 612enum ib_wr_opcode { 613 IB_WR_RDMA_WRITE, 614 IB_WR_RDMA_WRITE_WITH_IMM, 615 IB_WR_SEND, 616 IB_WR_SEND_WITH_IMM, 617 IB_WR_RDMA_READ, 618 IB_WR_ATOMIC_CMP_AND_SWP, 619 IB_WR_ATOMIC_FETCH_AND_ADD 620}; 621 622enum ib_send_flags { 623 IB_SEND_FENCE = 1, 624 IB_SEND_SIGNALED = (1<<1), 625 IB_SEND_SOLICITED = (1<<2), 626 IB_SEND_INLINE = (1<<3) 627}; 628 629struct ib_sge { 630 u64 addr; 631 u32 length; 632 u32 lkey; 633}; 634 635struct ib_send_wr { 636 struct ib_send_wr *next; 637 u64 wr_id; 638 struct ib_sge *sg_list; 639 int num_sge; 640 enum ib_wr_opcode opcode; 641 int send_flags; 642 __be32 imm_data; 643 union { 644 struct { 645 u64 remote_addr; 646 u32 rkey; 647 } rdma; 648 struct { 649 u64 remote_addr; 650 u64 compare_add; 651 u64 swap; 652 u32 rkey; 653 } atomic; 654 struct { 655 struct ib_ah *ah; 656 u32 remote_qpn; 657 u32 remote_qkey; 658 u16 pkey_index; /* valid for GSI only */ 659 u8 port_num; /* valid for DR SMPs on switch only */ 660 } ud; 661 } wr; 662}; 663 664struct ib_recv_wr { 665 struct ib_recv_wr *next; 666 u64 wr_id; 667 struct ib_sge *sg_list; 668 int num_sge; 669}; 670 671enum ib_access_flags { 672 IB_ACCESS_LOCAL_WRITE = 1, 673 IB_ACCESS_REMOTE_WRITE = (1<<1), 674 IB_ACCESS_REMOTE_READ = (1<<2), 675 IB_ACCESS_REMOTE_ATOMIC = (1<<3), 676 IB_ACCESS_MW_BIND = (1<<4) 677}; 678 679struct ib_phys_buf { 680 u64 addr; 681 u64 size; 682}; 683 684struct ib_mr_attr { 685 struct ib_pd *pd; 686 u64 device_virt_addr; 687 u64 size; 688 int mr_access_flags; 689 u32 lkey; 690 u32 rkey; 691}; 692 693enum ib_mr_rereg_flags { 694 IB_MR_REREG_TRANS = 1, 695 IB_MR_REREG_PD = (1<<1), 696 IB_MR_REREG_ACCESS = (1<<2) 697}; 698 699struct ib_mw_bind { 700 struct ib_mr *mr; 701 u64 wr_id; 702 u64 addr; 703 u32 length; 704 int send_flags; 705 int mw_access_flags; 706}; 707 708struct ib_fmr_attr { 709 int max_pages; 710 int max_maps; 711 u8 page_shift; 712}; 713 714/* 715 * XXX can this really be on 7 different lists at once? 716 * 717 */ 718struct ib_ucontext { 719 struct ib_device *device; 720 TAILQ_ENTRY(ib_ucontext) pd_list; 721 TAILQ_ENTRY(ib_ucontext) mr_list; 722 TAILQ_ENTRY(ib_ucontext) mw_list; 723 TAILQ_ENTRY(ib_ucontext) cq_list; 724 TAILQ_ENTRY(ib_ucontext) qp_list; 725 TAILQ_ENTRY(ib_ucontext) srq_list; 726 TAILQ_ENTRY(ib_ucontext) ah_list; 727 int closing; 728}; 729 730struct ib_uobject { 731 u64 user_handle; /* handle given to us by userspace */ 732 struct ib_ucontext *context; /* associated user context */ 733 void *object; /* containing object */ 734 TAILQ_ENTRY(ib_uobject) entry; /* link to context's list */ 735 u32 id; /* index into kernel idr */ 736 volatile uint32_t ref; 737 struct mtx lock; /* protects .live */ 738 int live; 739}; 740 741struct ib_udata { 742 void *inbuf; 743 void *outbuf; 744 size_t inlen; 745 size_t outlen; 746}; 747 748#define IB_UMEM_MAX_PAGE_CHUNK \ 749 ((PAGE_SIZE - offsetof(struct ib_umem_chunk, page_list)) / \ 750 ((void *) &((struct ib_umem_chunk *) 0)->page_list[1] - \ 751 (void *) &((struct ib_umem_chunk *) 0)->page_list[0])) 752 753struct ib_pd { 754 struct ib_device *device; 755 struct ib_uobject *uobject; 756 volatile int usecnt; /* count all resources */ 757}; 758 759struct ib_ah { 760 struct ib_device *device; 761 struct ib_pd *pd; 762 struct ib_uobject *uobject; 763}; 764 765typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context); 766 767struct ib_cq { 768 struct ib_device *device; 769 struct ib_uobject *uobject; 770 ib_comp_handler comp_handler; 771 void (*event_handler)(struct ib_event *, void *); 772 void * cq_context; 773 int cqe; 774 volatile int usecnt; /* count number of work queues */ 775}; 776 777struct ib_srq { 778 struct ib_device *device; 779 struct ib_pd *pd; 780 struct ib_uobject *uobject; 781 void (*event_handler)(struct ib_event *, void *); 782 void *srq_context; 783 volatile int usecnt; 784}; 785 786struct ib_qp { 787 struct ib_device *device; 788 struct ib_pd *pd; 789 struct ib_cq *send_cq; 790 struct ib_cq *recv_cq; 791 struct ib_srq *srq; 792 struct ib_uobject *uobject; 793 void (*event_handler)(struct ib_event *, void *); 794 void *qp_context; 795 u32 qp_num; 796 enum ib_qp_type qp_type; 797}; 798 799struct ib_mr { 800 struct ib_device *device; 801 struct ib_pd *pd; 802 struct ib_uobject *uobject; 803 u32 lkey; 804 u32 rkey; 805 volatile int usecnt; /* count number of MWs */ 806}; 807 808struct ib_mw { 809 struct ib_device *device; 810 struct ib_pd *pd; 811 struct ib_uobject *uobject; 812 u32 rkey; 813}; 814 815 816struct ib_fmr { 817 struct ib_device *device; 818 struct ib_pd *pd; 819 TAILQ_ENTRY(ib_fmr) entry; 820 u32 lkey; 821 u32 rkey; 822}; 823 824TAILQ_HEAD(ib_fmr_list_head, ib_fmr); 825 826struct ib_mad; 827struct ib_grh; 828 829enum ib_process_mad_flags { 830 IB_MAD_IGNORE_MKEY = 1, 831 IB_MAD_IGNORE_BKEY = 2, 832 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY 833}; 834 835enum ib_mad_result { 836 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */ 837 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */ 838 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */ 839 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */ 840}; 841 842#define IB_DEVICE_NAME_MAX 64 843 844struct ib_cache { 845 struct mtx lock; 846 struct ib_event_handler event_handler; 847 struct ib_pkey_cache **pkey_cache; 848 struct ib_gid_cache **gid_cache; 849 u8 *lmc_cache; 850}; 851 852struct ib_dma_mapping_ops { 853 int (*mapping_error)(struct ib_device *dev, 854 u64 dma_addr); 855 u64 (*map_single)(struct ib_device *dev, 856 void *ptr, size_t size, 857 enum dma_data_direction direction); 858 void (*unmap_single)(struct ib_device *dev, 859 u64 addr, size_t size, 860 enum dma_data_direction direction); 861 u64 (*map_page)(struct ib_device *dev, 862 void *page, unsigned long offset, 863 size_t size, 864 enum dma_data_direction direction); 865 void (*unmap_page)(struct ib_device *dev, 866 u64 addr, size_t size, 867 enum dma_data_direction direction); 868 int (*map_sg)(struct ib_device *dev, 869 struct rdma_scatterlist *sg, int nents, 870 enum dma_data_direction direction); 871 void (*unmap_sg)(struct ib_device *dev, 872 struct rdma_scatterlist *sg, int nents, 873 enum dma_data_direction direction); 874 u64 (*dma_address)(struct ib_device *dev, 875 struct rdma_scatterlist *sg); 876 unsigned int (*dma_len)(struct ib_device *dev, 877 struct rdma_scatterlist *sg); 878 void (*sync_single_for_cpu)(struct ib_device *dev, 879 u64 dma_handle, 880 size_t size, 881 enum dma_data_direction dir); 882 void (*sync_single_for_device)(struct ib_device *dev, 883 u64 dma_handle, 884 size_t size, 885 enum dma_data_direction dir); 886 void *(*alloc_coherent)(struct ib_device *dev, 887 size_t size, 888 u64 *dma_handle, 889 int flag); 890 void (*free_coherent)(struct ib_device *dev, 891 size_t size, void *cpu_addr, 892 u64 dma_handle); 893}; 894 895struct iw_cm_verbs; 896 897struct ib_device { 898 struct device *dma_device; 899 900 char name[IB_DEVICE_NAME_MAX]; 901 902 TAILQ_HEAD(, ib_event_handler) event_handler_list; 903 struct mtx event_handler_lock; 904 905 TAILQ_ENTRY(ib_device) core_list; 906 TAILQ_HEAD(, ib_client_data) client_data_list; 907 struct mtx client_data_lock; 908 909 struct ib_cache cache; 910 int *pkey_tbl_len; 911 int *gid_tbl_len; 912 913 u32 flags; 914 915 int num_comp_vectors; 916 917 struct iw_cm_verbs *iwcm; 918 919 int (*query_device)(struct ib_device *device, 920 struct ib_device_attr *device_attr); 921 int (*query_port)(struct ib_device *device, 922 u8 port_num, 923 struct ib_port_attr *port_attr); 924 int (*query_gid)(struct ib_device *device, 925 u8 port_num, int index, 926 union ib_gid *gid); 927 int (*query_pkey)(struct ib_device *device, 928 u8 port_num, u16 index, u16 *pkey); 929 int (*modify_device)(struct ib_device *device, 930 int device_modify_mask, 931 struct ib_device_modify *device_modify); 932 int (*modify_port)(struct ib_device *device, 933 u8 port_num, int port_modify_mask, 934 struct ib_port_modify *port_modify); 935 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device, 936 struct ib_udata *udata); 937 int (*dealloc_ucontext)(struct ib_ucontext *context); 938 int (*mmap)(struct ib_ucontext *context, 939 struct vm_object *vma); 940 struct ib_pd * (*alloc_pd)(struct ib_device *device, 941 struct ib_ucontext *context, 942 struct ib_udata *udata); 943 int (*dealloc_pd)(struct ib_pd *pd); 944 struct ib_ah * (*create_ah)(struct ib_pd *pd, 945 struct ib_ah_attr *ah_attr); 946 int (*modify_ah)(struct ib_ah *ah, 947 struct ib_ah_attr *ah_attr); 948 int (*query_ah)(struct ib_ah *ah, 949 struct ib_ah_attr *ah_attr); 950 int (*destroy_ah)(struct ib_ah *ah); 951 struct ib_srq * (*create_srq)(struct ib_pd *pd, 952 struct ib_srq_init_attr *srq_init_attr, 953 struct ib_udata *udata); 954 int (*modify_srq)(struct ib_srq *srq, 955 struct ib_srq_attr *srq_attr, 956 enum ib_srq_attr_mask srq_attr_mask, 957 struct ib_udata *udata); 958 int (*query_srq)(struct ib_srq *srq, 959 struct ib_srq_attr *srq_attr); 960 int (*destroy_srq)(struct ib_srq *srq); 961 int (*post_srq_recv)(struct ib_srq *srq, 962 struct ib_recv_wr *recv_wr, 963 struct ib_recv_wr **bad_recv_wr); 964 struct ib_qp * (*create_qp)(struct ib_pd *pd, 965 struct ib_qp_init_attr *qp_init_attr, 966 struct ib_udata *udata); 967 int (*modify_qp)(struct ib_qp *qp, 968 struct ib_qp_attr *qp_attr, 969 int qp_attr_mask, 970 struct ib_udata *udata); 971 int (*query_qp)(struct ib_qp *qp, 972 struct ib_qp_attr *qp_attr, 973 int qp_attr_mask, 974 struct ib_qp_init_attr *qp_init_attr); 975 int (*destroy_qp)(struct ib_qp *qp); 976 int (*post_send)(struct ib_qp *qp, 977 struct ib_send_wr *send_wr, 978 struct ib_send_wr **bad_send_wr); 979 int (*post_recv)(struct ib_qp *qp, 980 struct ib_recv_wr *recv_wr, 981 struct ib_recv_wr **bad_recv_wr); 982 struct ib_cq * (*create_cq)(struct ib_device *device, int cqe, 983 int comp_vector, 984 struct ib_ucontext *context, 985 struct ib_udata *udata); 986 int (*destroy_cq)(struct ib_cq *cq); 987 int (*resize_cq)(struct ib_cq *cq, int cqe, 988 struct ib_udata *udata); 989 int (*poll_cq)(struct ib_cq *cq, int num_entries, 990 struct ib_wc *wc); 991 int (*peek_cq)(struct ib_cq *cq, int wc_cnt); 992 int (*req_notify_cq)(struct ib_cq *cq, 993 enum ib_cq_notify_flags flags); 994 int (*req_ncomp_notif)(struct ib_cq *cq, 995 int wc_cnt); 996 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd, 997 int mr_access_flags); 998 struct ib_mr * (*reg_phys_mr)(struct ib_pd *pd, 999 struct ib_phys_buf *phys_buf_array, 1000 int num_phys_buf, 1001 int mr_access_flags, 1002 u64 *iova_start); 1003 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd, 1004 u64 start, u64 length, 1005 u64 virt_addr, 1006 int mr_access_flags, 1007 struct ib_udata *udata); 1008 int (*query_mr)(struct ib_mr *mr, 1009 struct ib_mr_attr *mr_attr); 1010 int (*dereg_mr)(struct ib_mr *mr); 1011 int (*rereg_phys_mr)(struct ib_mr *mr, 1012 int mr_rereg_mask, 1013 struct ib_pd *pd, 1014 struct ib_phys_buf *phys_buf_array, 1015 int num_phys_buf, 1016 int mr_access_flags, 1017 u64 *iova_start); 1018 struct ib_mw * (*alloc_mw)(struct ib_pd *pd); 1019 int (*bind_mw)(struct ib_qp *qp, 1020 struct ib_mw *mw, 1021 struct ib_mw_bind *mw_bind); 1022 int (*dealloc_mw)(struct ib_mw *mw); 1023 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd, 1024 int mr_access_flags, 1025 struct ib_fmr_attr *fmr_attr); 1026 int (*map_phys_fmr)(struct ib_fmr *fmr, 1027 u64 *page_list, int list_len, 1028 u64 iova); 1029 int (*unmap_fmr)(struct ib_fmr_list_head *fmr_list); 1030 int (*dealloc_fmr)(struct ib_fmr *fmr); 1031 int (*attach_mcast)(struct ib_qp *qp, 1032 union ib_gid *gid, 1033 u16 lid); 1034 int (*detach_mcast)(struct ib_qp *qp, 1035 union ib_gid *gid, 1036 u16 lid); 1037 int (*process_mad)(struct ib_device *device, 1038 int process_mad_flags, 1039 u8 port_num, 1040 struct ib_wc *in_wc, 1041 struct ib_grh *in_grh, 1042 struct ib_mad *in_mad, 1043 struct ib_mad *out_mad); 1044 1045 struct ib_dma_mapping_ops *dma_ops; 1046 1047 struct module *owner; 1048#ifdef notyet 1049 struct class_device class_dev; 1050 struct kobject ports_parent; 1051 struct list_head port_list; 1052#endif 1053 enum { 1054 IB_DEV_UNINITIALIZED, 1055 IB_DEV_REGISTERED, 1056 IB_DEV_UNREGISTERED 1057 } reg_state; 1058 1059 u64 uverbs_cmd_mask; 1060 int uverbs_abi_ver; 1061 1062 char node_desc[64]; 1063 __be64 node_guid; 1064 u8 node_type; 1065 u8 phys_port_cnt; 1066}; 1067 1068struct ib_client { 1069 char *name; 1070 void (*add) (struct ib_device *); 1071 void (*remove)(struct ib_device *); 1072 TAILQ_ENTRY(ib_client) list; 1073}; 1074 1075struct ib_device *ib_alloc_device(size_t size); 1076void ib_dealloc_device(struct ib_device *device); 1077 1078int ib_register_device (struct ib_device *device); 1079void ib_unregister_device(struct ib_device *device); 1080 1081int ib_register_client (struct ib_client *client); 1082void ib_unregister_client(struct ib_client *client); 1083 1084void *ib_get_client_data(struct ib_device *device, struct ib_client *client); 1085void ib_set_client_data(struct ib_device *device, struct ib_client *client, 1086 void *data); 1087 1088static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len) 1089{ 1090 return copyin(udata->inbuf, dest, len); 1091} 1092 1093static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len) 1094{ 1095 return copyout(src, udata->outbuf, len); 1096} 1097 1098/** 1099 * ib_modify_qp_is_ok - Check that the supplied attribute mask 1100 * contains all required attributes and no attributes not allowed for 1101 * the given QP state transition. 1102 * @cur_state: Current QP state 1103 * @next_state: Next QP state 1104 * @type: QP type 1105 * @mask: Mask of supplied QP attributes 1106 * 1107 * This function is a helper function that a low-level driver's 1108 * modify_qp method can use to validate the consumer's input. It 1109 * checks that cur_state and next_state are valid QP states, that a 1110 * transition from cur_state to next_state is allowed by the IB spec, 1111 * and that the attribute mask supplied is allowed for the transition. 1112 */ 1113int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state, 1114 enum ib_qp_type type, enum ib_qp_attr_mask mask); 1115 1116int ib_register_event_handler (struct ib_event_handler *event_handler); 1117int ib_unregister_event_handler(struct ib_event_handler *event_handler); 1118void ib_dispatch_event(struct ib_event *event); 1119 1120int ib_query_device(struct ib_device *device, 1121 struct ib_device_attr *device_attr); 1122 1123int ib_query_port(struct ib_device *device, 1124 u8 port_num, struct ib_port_attr *port_attr); 1125 1126int ib_query_gid(struct ib_device *device, 1127 u8 port_num, int index, union ib_gid *gid); 1128 1129int ib_query_pkey(struct ib_device *device, 1130 u8 port_num, u16 index, u16 *pkey); 1131 1132int ib_modify_device(struct ib_device *device, 1133 int device_modify_mask, 1134 struct ib_device_modify *device_modify); 1135 1136int ib_modify_port(struct ib_device *device, 1137 u8 port_num, int port_modify_mask, 1138 struct ib_port_modify *port_modify); 1139 1140int ib_find_gid(struct ib_device *device, union ib_gid *gid, 1141 u8 *port_num, u16 *index); 1142 1143int ib_find_pkey(struct ib_device *device, 1144 u8 port_num, u16 pkey, u16 *index); 1145 1146/** 1147 * ib_alloc_pd - Allocates an unused protection domain. 1148 * @device: The device on which to allocate the protection domain. 1149 * 1150 * A protection domain object provides an association between QPs, shared 1151 * receive queues, address handles, memory regions, and memory windows. 1152 */ 1153struct ib_pd *ib_alloc_pd(struct ib_device *device); 1154 1155/** 1156 * ib_dealloc_pd - Deallocates a protection domain. 1157 * @pd: The protection domain to deallocate. 1158 */ 1159int ib_dealloc_pd(struct ib_pd *pd); 1160 1161/** 1162 * ib_create_ah - Creates an address handle for the given address vector. 1163 * @pd: The protection domain associated with the address handle. 1164 * @ah_attr: The attributes of the address vector. 1165 * 1166 * The address handle is used to reference a local or global destination 1167 * in all UD QP post sends. 1168 */ 1169struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr); 1170 1171/** 1172 * ib_init_ah_from_wc - Initializes address handle attributes from a 1173 * work completion. 1174 * @device: Device on which the received message arrived. 1175 * @port_num: Port on which the received message arrived. 1176 * @wc: Work completion associated with the received message. 1177 * @grh: References the received global route header. This parameter is 1178 * ignored unless the work completion indicates that the GRH is valid. 1179 * @ah_attr: Returned attributes that can be used when creating an address 1180 * handle for replying to the message. 1181 */ 1182int ib_init_ah_from_wc(struct ib_device *device, u8 port_num, struct ib_wc *wc, 1183 struct ib_grh *grh, struct ib_ah_attr *ah_attr); 1184 1185/** 1186 * ib_create_ah_from_wc - Creates an address handle associated with the 1187 * sender of the specified work completion. 1188 * @pd: The protection domain associated with the address handle. 1189 * @wc: Work completion information associated with a received message. 1190 * @grh: References the received global route header. This parameter is 1191 * ignored unless the work completion indicates that the GRH is valid. 1192 * @port_num: The outbound port number to associate with the address. 1193 * 1194 * The address handle is used to reference a local or global destination 1195 * in all UD QP post sends. 1196 */ 1197struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, struct ib_wc *wc, 1198 struct ib_grh *grh, u8 port_num); 1199 1200/** 1201 * ib_modify_ah - Modifies the address vector associated with an address 1202 * handle. 1203 * @ah: The address handle to modify. 1204 * @ah_attr: The new address vector attributes to associate with the 1205 * address handle. 1206 */ 1207int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr); 1208 1209/** 1210 * ib_query_ah - Queries the address vector associated with an address 1211 * handle. 1212 * @ah: The address handle to query. 1213 * @ah_attr: The address vector attributes associated with the address 1214 * handle. 1215 */ 1216int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr); 1217 1218/** 1219 * ib_destroy_ah - Destroys an address handle. 1220 * @ah: The address handle to destroy. 1221 */ 1222int ib_destroy_ah(struct ib_ah *ah); 1223 1224/** 1225 * ib_create_srq - Creates a SRQ associated with the specified protection 1226 * domain. 1227 * @pd: The protection domain associated with the SRQ. 1228 * @srq_init_attr: A list of initial attributes required to create the 1229 * SRQ. If SRQ creation succeeds, then the attributes are updated to 1230 * the actual capabilities of the created SRQ. 1231 * 1232 * srq_attr->max_wr and srq_attr->max_sge are read the determine the 1233 * requested size of the SRQ, and set to the actual values allocated 1234 * on return. If ib_create_srq() succeeds, then max_wr and max_sge 1235 * will always be at least as large as the requested values. 1236 */ 1237struct ib_srq *ib_create_srq(struct ib_pd *pd, 1238 struct ib_srq_init_attr *srq_init_attr); 1239 1240/** 1241 * ib_modify_srq - Modifies the attributes for the specified SRQ. 1242 * @srq: The SRQ to modify. 1243 * @srq_attr: On input, specifies the SRQ attributes to modify. On output, 1244 * the current values of selected SRQ attributes are returned. 1245 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ 1246 * are being modified. 1247 * 1248 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or 1249 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when 1250 * the number of receives queued drops below the limit. 1251 */ 1252int ib_modify_srq(struct ib_srq *srq, 1253 struct ib_srq_attr *srq_attr, 1254 enum ib_srq_attr_mask srq_attr_mask); 1255 1256/** 1257 * ib_query_srq - Returns the attribute list and current values for the 1258 * specified SRQ. 1259 * @srq: The SRQ to query. 1260 * @srq_attr: The attributes of the specified SRQ. 1261 */ 1262int ib_query_srq(struct ib_srq *srq, 1263 struct ib_srq_attr *srq_attr); 1264 1265/** 1266 * ib_destroy_srq - Destroys the specified SRQ. 1267 * @srq: The SRQ to destroy. 1268 */ 1269int ib_destroy_srq(struct ib_srq *srq); 1270 1271/** 1272 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ. 1273 * @srq: The SRQ to post the work request on. 1274 * @recv_wr: A list of work requests to post on the receive queue. 1275 * @bad_recv_wr: On an immediate failure, this parameter will reference 1276 * the work request that failed to be posted on the QP. 1277 */ 1278static inline int ib_post_srq_recv(struct ib_srq *srq, 1279 struct ib_recv_wr *recv_wr, 1280 struct ib_recv_wr **bad_recv_wr) 1281{ 1282 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr); 1283} 1284 1285/** 1286 * ib_create_qp - Creates a QP associated with the specified protection 1287 * domain. 1288 * @pd: The protection domain associated with the QP. 1289 * @qp_init_attr: A list of initial attributes required to create the 1290 * QP. If QP creation succeeds, then the attributes are updated to 1291 * the actual capabilities of the created QP. 1292 */ 1293struct ib_qp *ib_create_qp(struct ib_pd *pd, 1294 struct ib_qp_init_attr *qp_init_attr); 1295 1296/** 1297 * ib_modify_qp - Modifies the attributes for the specified QP and then 1298 * transitions the QP to the given state. 1299 * @qp: The QP to modify. 1300 * @qp_attr: On input, specifies the QP attributes to modify. On output, 1301 * the current values of selected QP attributes are returned. 1302 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP 1303 * are being modified. 1304 */ 1305int ib_modify_qp(struct ib_qp *qp, 1306 struct ib_qp_attr *qp_attr, 1307 int qp_attr_mask); 1308 1309/** 1310 * ib_query_qp - Returns the attribute list and current values for the 1311 * specified QP. 1312 * @qp: The QP to query. 1313 * @qp_attr: The attributes of the specified QP. 1314 * @qp_attr_mask: A bit-mask used to select specific attributes to query. 1315 * @qp_init_attr: Additional attributes of the selected QP. 1316 * 1317 * The qp_attr_mask may be used to limit the query to gathering only the 1318 * selected attributes. 1319 */ 1320int ib_query_qp(struct ib_qp *qp, 1321 struct ib_qp_attr *qp_attr, 1322 int qp_attr_mask, 1323 struct ib_qp_init_attr *qp_init_attr); 1324 1325/** 1326 * ib_destroy_qp - Destroys the specified QP. 1327 * @qp: The QP to destroy. 1328 */ 1329int ib_destroy_qp(struct ib_qp *qp); 1330 1331/** 1332 * ib_post_send - Posts a list of work requests to the send queue of 1333 * the specified QP. 1334 * @qp: The QP to post the work request on. 1335 * @send_wr: A list of work requests to post on the send queue. 1336 * @bad_send_wr: On an immediate failure, this parameter will reference 1337 * the work request that failed to be posted on the QP. 1338 */ 1339static inline int ib_post_send(struct ib_qp *qp, 1340 struct ib_send_wr *send_wr, 1341 struct ib_send_wr **bad_send_wr) 1342{ 1343 return qp->device->post_send(qp, send_wr, bad_send_wr); 1344} 1345 1346/** 1347 * ib_post_recv - Posts a list of work requests to the receive queue of 1348 * the specified QP. 1349 * @qp: The QP to post the work request on. 1350 * @recv_wr: A list of work requests to post on the receive queue. 1351 * @bad_recv_wr: On an immediate failure, this parameter will reference 1352 * the work request that failed to be posted on the QP. 1353 */ 1354static inline int ib_post_recv(struct ib_qp *qp, 1355 struct ib_recv_wr *recv_wr, 1356 struct ib_recv_wr **bad_recv_wr) 1357{ 1358 return qp->device->post_recv(qp, recv_wr, bad_recv_wr); 1359} 1360 1361/** 1362 * ib_create_cq - Creates a CQ on the specified device. 1363 * @device: The device on which to create the CQ. 1364 * @comp_handler: A user-specified callback that is invoked when a 1365 * completion event occurs on the CQ. 1366 * @event_handler: A user-specified callback that is invoked when an 1367 * asynchronous event not associated with a completion occurs on the CQ. 1368 * @cq_context: Context associated with the CQ returned to the user via 1369 * the associated completion and event handlers. 1370 * @cqe: The minimum size of the CQ. 1371 * @comp_vector - Completion vector used to signal completion events. 1372 * Must be >= 0 and < context->num_comp_vectors. 1373 * 1374 * Users can examine the cq structure to determine the actual CQ size. 1375 */ 1376struct ib_cq *ib_create_cq(struct ib_device *device, 1377 ib_comp_handler comp_handler, 1378 void (*event_handler)(struct ib_event *, void *), 1379 void *cq_context, int cqe, int comp_vector); 1380 1381/** 1382 * ib_resize_cq - Modifies the capacity of the CQ. 1383 * @cq: The CQ to resize. 1384 * @cqe: The minimum size of the CQ. 1385 * 1386 * Users can examine the cq structure to determine the actual CQ size. 1387 */ 1388int ib_resize_cq(struct ib_cq *cq, int cqe); 1389 1390/** 1391 * ib_destroy_cq - Destroys the specified CQ. 1392 * @cq: The CQ to destroy. 1393 */ 1394int ib_destroy_cq(struct ib_cq *cq); 1395 1396/** 1397 * ib_poll_cq - poll a CQ for completion(s) 1398 * @cq:the CQ being polled 1399 * @num_entries:maximum number of completions to return 1400 * @wc:array of at least @num_entries &struct ib_wc where completions 1401 * will be returned 1402 * 1403 * Poll a CQ for (possibly multiple) completions. If the return value 1404 * is < 0, an error occurred. If the return value is >= 0, it is the 1405 * number of completions returned. If the return value is 1406 * non-negative and < num_entries, then the CQ was emptied. 1407 */ 1408static inline int ib_poll_cq(struct ib_cq *cq, int num_entries, 1409 struct ib_wc *wc) 1410{ 1411 return cq->device->poll_cq(cq, num_entries, wc); 1412} 1413 1414/** 1415 * ib_peek_cq - Returns the number of unreaped completions currently 1416 * on the specified CQ. 1417 * @cq: The CQ to peek. 1418 * @wc_cnt: A minimum number of unreaped completions to check for. 1419 * 1420 * If the number of unreaped completions is greater than or equal to wc_cnt, 1421 * this function returns wc_cnt, otherwise, it returns the actual number of 1422 * unreaped completions. 1423 */ 1424int ib_peek_cq(struct ib_cq *cq, int wc_cnt); 1425 1426/** 1427 * ib_req_notify_cq - Request completion notification on a CQ. 1428 * @cq: The CQ to generate an event for. 1429 * @flags: 1430 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP 1431 * to request an event on the next solicited event or next work 1432 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS 1433 * may also be |ed in to request a hint about missed events, as 1434 * described below. 1435 * 1436 * Return Value: 1437 * < 0 means an error occurred while requesting notification 1438 * == 0 means notification was requested successfully, and if 1439 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events 1440 * were missed and it is safe to wait for another event. In 1441 * this case is it guaranteed that any work completions added 1442 * to the CQ since the last CQ poll will trigger a completion 1443 * notification event. 1444 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed 1445 * in. It means that the consumer must poll the CQ again to 1446 * make sure it is empty to avoid missing an event because of a 1447 * race between requesting notification and an entry being 1448 * added to the CQ. This return value means it is possible 1449 * (but not guaranteed) that a work completion has been added 1450 * to the CQ since the last poll without triggering a 1451 * completion notification event. 1452 */ 1453static inline int ib_req_notify_cq(struct ib_cq *cq, 1454 enum ib_cq_notify_flags flags) 1455{ 1456 return cq->device->req_notify_cq(cq, flags); 1457} 1458 1459/** 1460 * ib_req_ncomp_notif - Request completion notification when there are 1461 * at least the specified number of unreaped completions on the CQ. 1462 * @cq: The CQ to generate an event for. 1463 * @wc_cnt: The number of unreaped completions that should be on the 1464 * CQ before an event is generated. 1465 */ 1466static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt) 1467{ 1468 return cq->device->req_ncomp_notif ? 1469 cq->device->req_ncomp_notif(cq, wc_cnt) : 1470 ENOSYS; 1471} 1472 1473/** 1474 * ib_get_dma_mr - Returns a memory region for system memory that is 1475 * usable for DMA. 1476 * @pd: The protection domain associated with the memory region. 1477 * @mr_access_flags: Specifies the memory access rights. 1478 * 1479 * Note that the ib_dma_*() functions defined below must be used 1480 * to create/destroy addresses used with the Lkey or Rkey returned 1481 * by ib_get_dma_mr(). 1482 */ 1483struct ib_mr *ib_get_dma_mr(struct ib_pd *pd, int mr_access_flags); 1484#ifdef notyet 1485/** 1486 * ib_dma_mapping_error - check a DMA addr for error 1487 * @dev: The device for which the dma_addr was created 1488 * @dma_addr: The DMA address to check 1489 */ 1490static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr) 1491{ 1492 if (dev->dma_ops) 1493 return dev->dma_ops->mapping_error(dev, dma_addr); 1494 return dma_mapping_error(dma_addr); 1495} 1496 1497/** 1498 * ib_dma_map_single - Map a kernel virtual address to DMA address 1499 * @dev: The device for which the dma_addr is to be created 1500 * @cpu_addr: The kernel virtual address 1501 * @size: The size of the region in bytes 1502 * @direction: The direction of the DMA 1503 */ 1504static inline u64 ib_dma_map_single(struct ib_device *dev, 1505 void *cpu_addr, size_t size, 1506 enum dma_data_direction direction) 1507{ 1508 if (dev->dma_ops) 1509 return dev->dma_ops->map_single(dev, cpu_addr, size, direction); 1510 return dma_map_single(dev->dma_device, cpu_addr, size, direction); 1511} 1512 1513/** 1514 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single() 1515 * @dev: The device for which the DMA address was created 1516 * @addr: The DMA address 1517 * @size: The size of the region in bytes 1518 * @direction: The direction of the DMA 1519 */ 1520static inline void ib_dma_unmap_single(struct ib_device *dev, 1521 u64 addr, size_t size, 1522 enum dma_data_direction direction) 1523{ 1524 if (dev->dma_ops) 1525 dev->dma_ops->unmap_single(dev, addr, size, direction); 1526 else 1527 dma_unmap_single(dev->dma_device, addr, size, direction); 1528} 1529 1530/** 1531 * ib_dma_map_page - Map a physical page to DMA address 1532 * @dev: The device for which the dma_addr is to be created 1533 * @page: The page to be mapped 1534 * @offset: The offset within the page 1535 * @size: The size of the region in bytes 1536 * @direction: The direction of the DMA 1537 */ 1538static inline u64 ib_dma_map_page(struct ib_device *dev, 1539 struct page *page, 1540 unsigned long offset, 1541 size_t size, 1542 enum dma_data_direction direction) 1543{ 1544 if (dev->dma_ops) 1545 return dev->dma_ops->map_page(dev, page, offset, size, direction); 1546 return dma_map_page(dev->dma_device, page, offset, size, direction); 1547} 1548 1549/** 1550 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page() 1551 * @dev: The device for which the DMA address was created 1552 * @addr: The DMA address 1553 * @size: The size of the region in bytes 1554 * @direction: The direction of the DMA 1555 */ 1556static inline void ib_dma_unmap_page(struct ib_device *dev, 1557 u64 addr, size_t size, 1558 enum dma_data_direction direction) 1559{ 1560 if (dev->dma_ops) 1561 dev->dma_ops->unmap_page(dev, addr, size, direction); 1562 else 1563 dma_unmap_page(dev->dma_device, addr, size, direction); 1564} 1565 1566/** 1567 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses 1568 * @dev: The device for which the DMA addresses are to be created 1569 * @sg: The array of scatter/gather entries 1570 * @nents: The number of scatter/gather entries 1571 * @direction: The direction of the DMA 1572 */ 1573static inline int ib_dma_map_sg(struct ib_device *dev, 1574 struct rdma_scatterlist *sg, int nents, 1575 enum dma_data_direction direction) 1576{ 1577 if (dev->dma_ops) 1578 return dev->dma_ops->map_sg(dev, sg, nents, direction); 1579 return dma_map_sg(dev->dma_device, sg, nents, direction); 1580} 1581 1582/** 1583 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses 1584 * @dev: The device for which the DMA addresses were created 1585 * @sg: The array of scatter/gather entries 1586 * @nents: The number of scatter/gather entries 1587 * @direction: The direction of the DMA 1588 */ 1589static inline void ib_dma_unmap_sg(struct ib_device *dev, 1590 struct rdma_scatterlist *sg, int nents, 1591 enum dma_data_direction direction) 1592{ 1593 if (dev->dma_ops) 1594 dev->dma_ops->unmap_sg(dev, sg, nents, direction); 1595 else 1596 dma_unmap_sg(dev->dma_device, sg, nents, direction); 1597} 1598 1599/** 1600 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry 1601 * @dev: The device for which the DMA addresses were created 1602 * @sg: The scatter/gather entry 1603 */ 1604static inline u64 ib_sg_dma_address(struct ib_device *dev, 1605 struct rdma_scatterlist *sg) 1606{ 1607 if (dev->dma_ops) 1608 return dev->dma_ops->dma_address(dev, sg); 1609 return sg_dma_address(sg); 1610} 1611 1612/** 1613 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry 1614 * @dev: The device for which the DMA addresses were created 1615 * @sg: The scatter/gather entry 1616 */ 1617static inline unsigned int ib_sg_dma_len(struct ib_device *dev, 1618 struct rdma_scatterlist *sg) 1619{ 1620 if (dev->dma_ops) 1621 return dev->dma_ops->dma_len(dev, sg); 1622 return sg_dma_len(sg); 1623} 1624 1625/** 1626 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU 1627 * @dev: The device for which the DMA address was created 1628 * @addr: The DMA address 1629 * @size: The size of the region in bytes 1630 * @dir: The direction of the DMA 1631 */ 1632static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev, 1633 u64 addr, 1634 size_t size, 1635 enum dma_data_direction dir) 1636{ 1637 if (dev->dma_ops) 1638 dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir); 1639 else 1640 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir); 1641} 1642 1643/** 1644 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device 1645 * @dev: The device for which the DMA address was created 1646 * @addr: The DMA address 1647 * @size: The size of the region in bytes 1648 * @dir: The direction of the DMA 1649 */ 1650static inline void ib_dma_sync_single_for_device(struct ib_device *dev, 1651 u64 addr, 1652 size_t size, 1653 enum dma_data_direction dir) 1654{ 1655 if (dev->dma_ops) 1656 dev->dma_ops->sync_single_for_device(dev, addr, size, dir); 1657 else 1658 dma_sync_single_for_device(dev->dma_device, addr, size, dir); 1659} 1660 1661/** 1662 * ib_dma_alloc_coherent - Allocate memory and map it for DMA 1663 * @dev: The device for which the DMA address is requested 1664 * @size: The size of the region to allocate in bytes 1665 * @dma_handle: A pointer for returning the DMA address of the region 1666 * @flag: memory allocator flags 1667 */ 1668static inline void *ib_dma_alloc_coherent(struct ib_device *dev, 1669 size_t size, 1670 u64 *dma_handle, 1671 gfp_t flag) 1672{ 1673 if (dev->dma_ops) 1674 return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag); 1675 else { 1676 dma_addr_t handle; 1677 void *ret; 1678 1679 ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag); 1680 *dma_handle = handle; 1681 return ret; 1682 } 1683} 1684 1685/** 1686 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent() 1687 * @dev: The device for which the DMA addresses were allocated 1688 * @size: The size of the region 1689 * @cpu_addr: the address returned by ib_dma_alloc_coherent() 1690 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent() 1691 */ 1692static inline void ib_dma_free_coherent(struct ib_device *dev, 1693 size_t size, void *cpu_addr, 1694 u64 dma_handle) 1695{ 1696 if (dev->dma_ops) 1697 dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle); 1698 else 1699 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle); 1700} 1701#endif 1702/** 1703 * ib_reg_phys_mr - Prepares a virtually addressed memory region for use 1704 * by an HCA. 1705 * @pd: The protection domain associated assigned to the registered region. 1706 * @phys_buf_array: Specifies a list of physical buffers to use in the 1707 * memory region. 1708 * @num_phys_buf: Specifies the size of the phys_buf_array. 1709 * @mr_access_flags: Specifies the memory access rights. 1710 * @iova_start: The offset of the region's starting I/O virtual address. 1711 */ 1712struct ib_mr *ib_reg_phys_mr(struct ib_pd *pd, 1713 struct ib_phys_buf *phys_buf_array, 1714 int num_phys_buf, 1715 int mr_access_flags, 1716 u64 *iova_start); 1717 1718/** 1719 * ib_rereg_phys_mr - Modifies the attributes of an existing memory region. 1720 * Conceptually, this call performs the functions deregister memory region 1721 * followed by register physical memory region. Where possible, 1722 * resources are reused instead of deallocated and reallocated. 1723 * @mr: The memory region to modify. 1724 * @mr_rereg_mask: A bit-mask used to indicate which of the following 1725 * properties of the memory region are being modified. 1726 * @pd: If %IB_MR_REREG_PD is set in mr_rereg_mask, this field specifies 1727 * the new protection domain to associated with the memory region, 1728 * otherwise, this parameter is ignored. 1729 * @phys_buf_array: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this 1730 * field specifies a list of physical buffers to use in the new 1731 * translation, otherwise, this parameter is ignored. 1732 * @num_phys_buf: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this 1733 * field specifies the size of the phys_buf_array, otherwise, this 1734 * parameter is ignored. 1735 * @mr_access_flags: If %IB_MR_REREG_ACCESS is set in mr_rereg_mask, this 1736 * field specifies the new memory access rights, otherwise, this 1737 * parameter is ignored. 1738 * @iova_start: The offset of the region's starting I/O virtual address. 1739 */ 1740int ib_rereg_phys_mr(struct ib_mr *mr, 1741 int mr_rereg_mask, 1742 struct ib_pd *pd, 1743 struct ib_phys_buf *phys_buf_array, 1744 int num_phys_buf, 1745 int mr_access_flags, 1746 u64 *iova_start); 1747 1748/** 1749 * ib_query_mr - Retrieves information about a specific memory region. 1750 * @mr: The memory region to retrieve information about. 1751 * @mr_attr: The attributes of the specified memory region. 1752 */ 1753int ib_query_mr(struct ib_mr *mr, struct ib_mr_attr *mr_attr); 1754 1755/** 1756 * ib_dereg_mr - Deregisters a memory region and removes it from the 1757 * HCA translation table. 1758 * @mr: The memory region to deregister. 1759 */ 1760int ib_dereg_mr(struct ib_mr *mr); 1761 1762/** 1763 * ib_alloc_mw - Allocates a memory window. 1764 * @pd: The protection domain associated with the memory window. 1765 */ 1766struct ib_mw *ib_alloc_mw(struct ib_pd *pd); 1767 1768/** 1769 * ib_bind_mw - Posts a work request to the send queue of the specified 1770 * QP, which binds the memory window to the given address range and 1771 * remote access attributes. 1772 * @qp: QP to post the bind work request on. 1773 * @mw: The memory window to bind. 1774 * @mw_bind: Specifies information about the memory window, including 1775 * its address range, remote access rights, and associated memory region. 1776 */ 1777static inline int ib_bind_mw(struct ib_qp *qp, 1778 struct ib_mw *mw, 1779 struct ib_mw_bind *mw_bind) 1780{ 1781 /* XXX reference counting in corresponding MR? */ 1782 return mw->device->bind_mw ? 1783 mw->device->bind_mw(qp, mw, mw_bind) : 1784 ENOSYS; 1785} 1786 1787/** 1788 * ib_dealloc_mw - Deallocates a memory window. 1789 * @mw: The memory window to deallocate. 1790 */ 1791int ib_dealloc_mw(struct ib_mw *mw); 1792 1793/** 1794 * ib_alloc_fmr - Allocates a unmapped fast memory region. 1795 * @pd: The protection domain associated with the unmapped region. 1796 * @mr_access_flags: Specifies the memory access rights. 1797 * @fmr_attr: Attributes of the unmapped region. 1798 * 1799 * A fast memory region must be mapped before it can be used as part of 1800 * a work request. 1801 */ 1802struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd, 1803 int mr_access_flags, 1804 struct ib_fmr_attr *fmr_attr); 1805 1806/** 1807 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region. 1808 * @fmr: The fast memory region to associate with the pages. 1809 * @page_list: An array of physical pages to map to the fast memory region. 1810 * @list_len: The number of pages in page_list. 1811 * @iova: The I/O virtual address to use with the mapped region. 1812 */ 1813static inline int ib_map_phys_fmr(struct ib_fmr *fmr, 1814 u64 *page_list, int list_len, 1815 u64 iova) 1816{ 1817 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova); 1818} 1819 1820/** 1821 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions. 1822 * @fmr_list: A linked list of fast memory regions to unmap. 1823 */ 1824int ib_unmap_fmr(struct ib_fmr_list_head *fmr_list); 1825 1826/** 1827 * ib_dealloc_fmr - Deallocates a fast memory region. 1828 * @fmr: The fast memory region to deallocate. 1829 */ 1830int ib_dealloc_fmr(struct ib_fmr *fmr); 1831 1832/** 1833 * ib_attach_mcast - Attaches the specified QP to a multicast group. 1834 * @qp: QP to attach to the multicast group. The QP must be type 1835 * IB_QPT_UD. 1836 * @gid: Multicast group GID. 1837 * @lid: Multicast group LID in host byte order. 1838 * 1839 * In order to send and receive multicast packets, subnet 1840 * administration must have created the multicast group and configured 1841 * the fabric appropriately. The port associated with the specified 1842 * QP must also be a member of the multicast group. 1843 */ 1844int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid); 1845 1846/** 1847 * ib_detach_mcast - Detaches the specified QP from a multicast group. 1848 * @qp: QP to detach from the multicast group. 1849 * @gid: Multicast group GID. 1850 * @lid: Multicast group LID in host byte order. 1851 */ 1852int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid); 1853 1854#endif /* IB_VERBS_H */ 1855