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 <uapi/linux/if_ether.h> 52 53 #include <linux/atomic.h> 54 #include <asm/uaccess.h> 55 56 extern struct workqueue_struct *ib_wq; 57 58 union ib_gid { 59 u8 raw[16]; 60 struct { 61 __be64 subnet_prefix; 62 __be64 interface_id; 63 } global; 64 }; 65 66 enum 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 RDMA_NODE_USNIC, 73 RDMA_NODE_USNIC_UDP, 74 }; 75 76 enum rdma_transport_type { 77 RDMA_TRANSPORT_IB, 78 RDMA_TRANSPORT_IWARP, 79 RDMA_TRANSPORT_USNIC, 80 RDMA_TRANSPORT_USNIC_UDP 81 }; 82 83 enum rdma_transport_type 84 rdma_node_get_transport(enum rdma_node_type node_type) __attribute_const__; 85 86 enum rdma_link_layer { 87 IB_LINK_LAYER_UNSPECIFIED, 88 IB_LINK_LAYER_INFINIBAND, 89 IB_LINK_LAYER_ETHERNET, 90 }; 91 92 enum ib_device_cap_flags { 93 IB_DEVICE_RESIZE_MAX_WR = 1, 94 IB_DEVICE_BAD_PKEY_CNTR = (1<<1), 95 IB_DEVICE_BAD_QKEY_CNTR = (1<<2), 96 IB_DEVICE_RAW_MULTI = (1<<3), 97 IB_DEVICE_AUTO_PATH_MIG = (1<<4), 98 IB_DEVICE_CHANGE_PHY_PORT = (1<<5), 99 IB_DEVICE_UD_AV_PORT_ENFORCE = (1<<6), 100 IB_DEVICE_CURR_QP_STATE_MOD = (1<<7), 101 IB_DEVICE_SHUTDOWN_PORT = (1<<8), 102 IB_DEVICE_INIT_TYPE = (1<<9), 103 IB_DEVICE_PORT_ACTIVE_EVENT = (1<<10), 104 IB_DEVICE_SYS_IMAGE_GUID = (1<<11), 105 IB_DEVICE_RC_RNR_NAK_GEN = (1<<12), 106 IB_DEVICE_SRQ_RESIZE = (1<<13), 107 IB_DEVICE_N_NOTIFY_CQ = (1<<14), 108 IB_DEVICE_LOCAL_DMA_LKEY = (1<<15), 109 IB_DEVICE_RESERVED = (1<<16), /* old SEND_W_INV */ 110 IB_DEVICE_MEM_WINDOW = (1<<17), 111 /* 112 * Devices should set IB_DEVICE_UD_IP_SUM if they support 113 * insertion of UDP and TCP checksum on outgoing UD IPoIB 114 * messages and can verify the validity of checksum for 115 * incoming messages. Setting this flag implies that the 116 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode. 117 */ 118 IB_DEVICE_UD_IP_CSUM = (1<<18), 119 IB_DEVICE_UD_TSO = (1<<19), 120 IB_DEVICE_XRC = (1<<20), 121 IB_DEVICE_MEM_MGT_EXTENSIONS = (1<<21), 122 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1<<22), 123 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1<<23), 124 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1<<24), 125 IB_DEVICE_MANAGED_FLOW_STEERING = (1<<29) 126 }; 127 128 enum ib_atomic_cap { 129 IB_ATOMIC_NONE, 130 IB_ATOMIC_HCA, 131 IB_ATOMIC_GLOB 132 }; 133 134 struct ib_device_attr { 135 u64 fw_ver; 136 __be64 sys_image_guid; 137 u64 max_mr_size; 138 u64 page_size_cap; 139 u32 vendor_id; 140 u32 vendor_part_id; 141 u32 hw_ver; 142 int max_qp; 143 int max_qp_wr; 144 int device_cap_flags; 145 int max_sge; 146 int max_sge_rd; 147 int max_cq; 148 int max_cqe; 149 int max_mr; 150 int max_pd; 151 int max_qp_rd_atom; 152 int max_ee_rd_atom; 153 int max_res_rd_atom; 154 int max_qp_init_rd_atom; 155 int max_ee_init_rd_atom; 156 enum ib_atomic_cap atomic_cap; 157 enum ib_atomic_cap masked_atomic_cap; 158 int max_ee; 159 int max_rdd; 160 int max_mw; 161 int max_raw_ipv6_qp; 162 int max_raw_ethy_qp; 163 int max_mcast_grp; 164 int max_mcast_qp_attach; 165 int max_total_mcast_qp_attach; 166 int max_ah; 167 int max_fmr; 168 int max_map_per_fmr; 169 int max_srq; 170 int max_srq_wr; 171 int max_srq_sge; 172 unsigned int max_fast_reg_page_list_len; 173 u16 max_pkeys; 174 u8 local_ca_ack_delay; 175 }; 176 177 enum ib_mtu { 178 IB_MTU_256 = 1, 179 IB_MTU_512 = 2, 180 IB_MTU_1024 = 3, 181 IB_MTU_2048 = 4, 182 IB_MTU_4096 = 5 183 }; 184 185 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu) 186 { 187 switch (mtu) { 188 case IB_MTU_256: return 256; 189 case IB_MTU_512: return 512; 190 case IB_MTU_1024: return 1024; 191 case IB_MTU_2048: return 2048; 192 case IB_MTU_4096: return 4096; 193 default: return -1; 194 } 195 } 196 197 enum ib_port_state { 198 IB_PORT_NOP = 0, 199 IB_PORT_DOWN = 1, 200 IB_PORT_INIT = 2, 201 IB_PORT_ARMED = 3, 202 IB_PORT_ACTIVE = 4, 203 IB_PORT_ACTIVE_DEFER = 5 204 }; 205 206 enum ib_port_cap_flags { 207 IB_PORT_SM = 1 << 1, 208 IB_PORT_NOTICE_SUP = 1 << 2, 209 IB_PORT_TRAP_SUP = 1 << 3, 210 IB_PORT_OPT_IPD_SUP = 1 << 4, 211 IB_PORT_AUTO_MIGR_SUP = 1 << 5, 212 IB_PORT_SL_MAP_SUP = 1 << 6, 213 IB_PORT_MKEY_NVRAM = 1 << 7, 214 IB_PORT_PKEY_NVRAM = 1 << 8, 215 IB_PORT_LED_INFO_SUP = 1 << 9, 216 IB_PORT_SM_DISABLED = 1 << 10, 217 IB_PORT_SYS_IMAGE_GUID_SUP = 1 << 11, 218 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP = 1 << 12, 219 IB_PORT_EXTENDED_SPEEDS_SUP = 1 << 14, 220 IB_PORT_CM_SUP = 1 << 16, 221 IB_PORT_SNMP_TUNNEL_SUP = 1 << 17, 222 IB_PORT_REINIT_SUP = 1 << 18, 223 IB_PORT_DEVICE_MGMT_SUP = 1 << 19, 224 IB_PORT_VENDOR_CLASS_SUP = 1 << 20, 225 IB_PORT_DR_NOTICE_SUP = 1 << 21, 226 IB_PORT_CAP_MASK_NOTICE_SUP = 1 << 22, 227 IB_PORT_BOOT_MGMT_SUP = 1 << 23, 228 IB_PORT_LINK_LATENCY_SUP = 1 << 24, 229 IB_PORT_CLIENT_REG_SUP = 1 << 25, 230 IB_PORT_IP_BASED_GIDS = 1 << 26 231 }; 232 233 enum ib_port_width { 234 IB_WIDTH_1X = 1, 235 IB_WIDTH_4X = 2, 236 IB_WIDTH_8X = 4, 237 IB_WIDTH_12X = 8 238 }; 239 240 static inline int ib_width_enum_to_int(enum ib_port_width width) 241 { 242 switch (width) { 243 case IB_WIDTH_1X: return 1; 244 case IB_WIDTH_4X: return 4; 245 case IB_WIDTH_8X: return 8; 246 case IB_WIDTH_12X: return 12; 247 default: return -1; 248 } 249 } 250 251 enum ib_port_speed { 252 IB_SPEED_SDR = 1, 253 IB_SPEED_DDR = 2, 254 IB_SPEED_QDR = 4, 255 IB_SPEED_FDR10 = 8, 256 IB_SPEED_FDR = 16, 257 IB_SPEED_EDR = 32 258 }; 259 260 struct ib_protocol_stats { 261 /* TBD... */ 262 }; 263 264 struct iw_protocol_stats { 265 u64 ipInReceives; 266 u64 ipInHdrErrors; 267 u64 ipInTooBigErrors; 268 u64 ipInNoRoutes; 269 u64 ipInAddrErrors; 270 u64 ipInUnknownProtos; 271 u64 ipInTruncatedPkts; 272 u64 ipInDiscards; 273 u64 ipInDelivers; 274 u64 ipOutForwDatagrams; 275 u64 ipOutRequests; 276 u64 ipOutDiscards; 277 u64 ipOutNoRoutes; 278 u64 ipReasmTimeout; 279 u64 ipReasmReqds; 280 u64 ipReasmOKs; 281 u64 ipReasmFails; 282 u64 ipFragOKs; 283 u64 ipFragFails; 284 u64 ipFragCreates; 285 u64 ipInMcastPkts; 286 u64 ipOutMcastPkts; 287 u64 ipInBcastPkts; 288 u64 ipOutBcastPkts; 289 290 u64 tcpRtoAlgorithm; 291 u64 tcpRtoMin; 292 u64 tcpRtoMax; 293 u64 tcpMaxConn; 294 u64 tcpActiveOpens; 295 u64 tcpPassiveOpens; 296 u64 tcpAttemptFails; 297 u64 tcpEstabResets; 298 u64 tcpCurrEstab; 299 u64 tcpInSegs; 300 u64 tcpOutSegs; 301 u64 tcpRetransSegs; 302 u64 tcpInErrs; 303 u64 tcpOutRsts; 304 }; 305 306 union rdma_protocol_stats { 307 struct ib_protocol_stats ib; 308 struct iw_protocol_stats iw; 309 }; 310 311 struct ib_port_attr { 312 enum ib_port_state state; 313 enum ib_mtu max_mtu; 314 enum ib_mtu active_mtu; 315 int gid_tbl_len; 316 u32 port_cap_flags; 317 u32 max_msg_sz; 318 u32 bad_pkey_cntr; 319 u32 qkey_viol_cntr; 320 u16 pkey_tbl_len; 321 u16 lid; 322 u16 sm_lid; 323 u8 lmc; 324 u8 max_vl_num; 325 u8 sm_sl; 326 u8 subnet_timeout; 327 u8 init_type_reply; 328 u8 active_width; 329 u8 active_speed; 330 u8 phys_state; 331 }; 332 333 enum ib_device_modify_flags { 334 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0, 335 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1 336 }; 337 338 struct ib_device_modify { 339 u64 sys_image_guid; 340 char node_desc[64]; 341 }; 342 343 enum ib_port_modify_flags { 344 IB_PORT_SHUTDOWN = 1, 345 IB_PORT_INIT_TYPE = (1<<2), 346 IB_PORT_RESET_QKEY_CNTR = (1<<3) 347 }; 348 349 struct ib_port_modify { 350 u32 set_port_cap_mask; 351 u32 clr_port_cap_mask; 352 u8 init_type; 353 }; 354 355 enum ib_event_type { 356 IB_EVENT_CQ_ERR, 357 IB_EVENT_QP_FATAL, 358 IB_EVENT_QP_REQ_ERR, 359 IB_EVENT_QP_ACCESS_ERR, 360 IB_EVENT_COMM_EST, 361 IB_EVENT_SQ_DRAINED, 362 IB_EVENT_PATH_MIG, 363 IB_EVENT_PATH_MIG_ERR, 364 IB_EVENT_DEVICE_FATAL, 365 IB_EVENT_PORT_ACTIVE, 366 IB_EVENT_PORT_ERR, 367 IB_EVENT_LID_CHANGE, 368 IB_EVENT_PKEY_CHANGE, 369 IB_EVENT_SM_CHANGE, 370 IB_EVENT_SRQ_ERR, 371 IB_EVENT_SRQ_LIMIT_REACHED, 372 IB_EVENT_QP_LAST_WQE_REACHED, 373 IB_EVENT_CLIENT_REREGISTER, 374 IB_EVENT_GID_CHANGE, 375 }; 376 377 struct ib_event { 378 struct ib_device *device; 379 union { 380 struct ib_cq *cq; 381 struct ib_qp *qp; 382 struct ib_srq *srq; 383 u8 port_num; 384 } element; 385 enum ib_event_type event; 386 }; 387 388 struct ib_event_handler { 389 struct ib_device *device; 390 void (*handler)(struct ib_event_handler *, struct ib_event *); 391 struct list_head list; 392 }; 393 394 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \ 395 do { \ 396 (_ptr)->device = _device; \ 397 (_ptr)->handler = _handler; \ 398 INIT_LIST_HEAD(&(_ptr)->list); \ 399 } while (0) 400 401 struct ib_global_route { 402 union ib_gid dgid; 403 u32 flow_label; 404 u8 sgid_index; 405 u8 hop_limit; 406 u8 traffic_class; 407 }; 408 409 struct ib_grh { 410 __be32 version_tclass_flow; 411 __be16 paylen; 412 u8 next_hdr; 413 u8 hop_limit; 414 union ib_gid sgid; 415 union ib_gid dgid; 416 }; 417 418 enum { 419 IB_MULTICAST_QPN = 0xffffff 420 }; 421 422 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF) 423 424 enum ib_ah_flags { 425 IB_AH_GRH = 1 426 }; 427 428 enum ib_rate { 429 IB_RATE_PORT_CURRENT = 0, 430 IB_RATE_2_5_GBPS = 2, 431 IB_RATE_5_GBPS = 5, 432 IB_RATE_10_GBPS = 3, 433 IB_RATE_20_GBPS = 6, 434 IB_RATE_30_GBPS = 4, 435 IB_RATE_40_GBPS = 7, 436 IB_RATE_60_GBPS = 8, 437 IB_RATE_80_GBPS = 9, 438 IB_RATE_120_GBPS = 10, 439 IB_RATE_14_GBPS = 11, 440 IB_RATE_56_GBPS = 12, 441 IB_RATE_112_GBPS = 13, 442 IB_RATE_168_GBPS = 14, 443 IB_RATE_25_GBPS = 15, 444 IB_RATE_100_GBPS = 16, 445 IB_RATE_200_GBPS = 17, 446 IB_RATE_300_GBPS = 18 447 }; 448 449 /** 450 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the 451 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be 452 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec. 453 * @rate: rate to convert. 454 */ 455 int ib_rate_to_mult(enum ib_rate rate) __attribute_const__; 456 457 /** 458 * ib_rate_to_mbps - Convert the IB rate enum to Mbps. 459 * For example, IB_RATE_2_5_GBPS will be converted to 2500. 460 * @rate: rate to convert. 461 */ 462 int ib_rate_to_mbps(enum ib_rate rate) __attribute_const__; 463 464 /** 465 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate 466 * enum. 467 * @mult: multiple to convert. 468 */ 469 enum ib_rate mult_to_ib_rate(int mult) __attribute_const__; 470 471 struct ib_ah_attr { 472 struct ib_global_route grh; 473 u16 dlid; 474 u8 sl; 475 u8 src_path_bits; 476 u8 static_rate; 477 u8 ah_flags; 478 u8 port_num; 479 u8 dmac[ETH_ALEN]; 480 u16 vlan_id; 481 }; 482 483 enum ib_wc_status { 484 IB_WC_SUCCESS, 485 IB_WC_LOC_LEN_ERR, 486 IB_WC_LOC_QP_OP_ERR, 487 IB_WC_LOC_EEC_OP_ERR, 488 IB_WC_LOC_PROT_ERR, 489 IB_WC_WR_FLUSH_ERR, 490 IB_WC_MW_BIND_ERR, 491 IB_WC_BAD_RESP_ERR, 492 IB_WC_LOC_ACCESS_ERR, 493 IB_WC_REM_INV_REQ_ERR, 494 IB_WC_REM_ACCESS_ERR, 495 IB_WC_REM_OP_ERR, 496 IB_WC_RETRY_EXC_ERR, 497 IB_WC_RNR_RETRY_EXC_ERR, 498 IB_WC_LOC_RDD_VIOL_ERR, 499 IB_WC_REM_INV_RD_REQ_ERR, 500 IB_WC_REM_ABORT_ERR, 501 IB_WC_INV_EECN_ERR, 502 IB_WC_INV_EEC_STATE_ERR, 503 IB_WC_FATAL_ERR, 504 IB_WC_RESP_TIMEOUT_ERR, 505 IB_WC_GENERAL_ERR 506 }; 507 508 enum ib_wc_opcode { 509 IB_WC_SEND, 510 IB_WC_RDMA_WRITE, 511 IB_WC_RDMA_READ, 512 IB_WC_COMP_SWAP, 513 IB_WC_FETCH_ADD, 514 IB_WC_BIND_MW, 515 IB_WC_LSO, 516 IB_WC_LOCAL_INV, 517 IB_WC_FAST_REG_MR, 518 IB_WC_MASKED_COMP_SWAP, 519 IB_WC_MASKED_FETCH_ADD, 520 /* 521 * Set value of IB_WC_RECV so consumers can test if a completion is a 522 * receive by testing (opcode & IB_WC_RECV). 523 */ 524 IB_WC_RECV = 1 << 7, 525 IB_WC_RECV_RDMA_WITH_IMM 526 }; 527 528 enum ib_wc_flags { 529 IB_WC_GRH = 1, 530 IB_WC_WITH_IMM = (1<<1), 531 IB_WC_WITH_INVALIDATE = (1<<2), 532 IB_WC_IP_CSUM_OK = (1<<3), 533 IB_WC_WITH_SMAC = (1<<4), 534 IB_WC_WITH_VLAN = (1<<5), 535 }; 536 537 struct ib_wc { 538 u64 wr_id; 539 enum ib_wc_status status; 540 enum ib_wc_opcode opcode; 541 u32 vendor_err; 542 u32 byte_len; 543 struct ib_qp *qp; 544 union { 545 __be32 imm_data; 546 u32 invalidate_rkey; 547 } ex; 548 u32 src_qp; 549 int wc_flags; 550 u16 pkey_index; 551 u16 slid; 552 u8 sl; 553 u8 dlid_path_bits; 554 u8 port_num; /* valid only for DR SMPs on switches */ 555 u8 smac[ETH_ALEN]; 556 u16 vlan_id; 557 }; 558 559 enum ib_cq_notify_flags { 560 IB_CQ_SOLICITED = 1 << 0, 561 IB_CQ_NEXT_COMP = 1 << 1, 562 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP, 563 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2, 564 }; 565 566 enum ib_srq_type { 567 IB_SRQT_BASIC, 568 IB_SRQT_XRC 569 }; 570 571 enum ib_srq_attr_mask { 572 IB_SRQ_MAX_WR = 1 << 0, 573 IB_SRQ_LIMIT = 1 << 1, 574 }; 575 576 struct ib_srq_attr { 577 u32 max_wr; 578 u32 max_sge; 579 u32 srq_limit; 580 }; 581 582 struct ib_srq_init_attr { 583 void (*event_handler)(struct ib_event *, void *); 584 void *srq_context; 585 struct ib_srq_attr attr; 586 enum ib_srq_type srq_type; 587 588 union { 589 struct { 590 struct ib_xrcd *xrcd; 591 struct ib_cq *cq; 592 } xrc; 593 } ext; 594 }; 595 596 struct ib_qp_cap { 597 u32 max_send_wr; 598 u32 max_recv_wr; 599 u32 max_send_sge; 600 u32 max_recv_sge; 601 u32 max_inline_data; 602 }; 603 604 enum ib_sig_type { 605 IB_SIGNAL_ALL_WR, 606 IB_SIGNAL_REQ_WR 607 }; 608 609 enum ib_qp_type { 610 /* 611 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries 612 * here (and in that order) since the MAD layer uses them as 613 * indices into a 2-entry table. 614 */ 615 IB_QPT_SMI, 616 IB_QPT_GSI, 617 618 IB_QPT_RC, 619 IB_QPT_UC, 620 IB_QPT_UD, 621 IB_QPT_RAW_IPV6, 622 IB_QPT_RAW_ETHERTYPE, 623 IB_QPT_RAW_PACKET = 8, 624 IB_QPT_XRC_INI = 9, 625 IB_QPT_XRC_TGT, 626 IB_QPT_MAX, 627 /* Reserve a range for qp types internal to the low level driver. 628 * These qp types will not be visible at the IB core layer, so the 629 * IB_QPT_MAX usages should not be affected in the core layer 630 */ 631 IB_QPT_RESERVED1 = 0x1000, 632 IB_QPT_RESERVED2, 633 IB_QPT_RESERVED3, 634 IB_QPT_RESERVED4, 635 IB_QPT_RESERVED5, 636 IB_QPT_RESERVED6, 637 IB_QPT_RESERVED7, 638 IB_QPT_RESERVED8, 639 IB_QPT_RESERVED9, 640 IB_QPT_RESERVED10, 641 }; 642 643 enum ib_qp_create_flags { 644 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0, 645 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1, 646 IB_QP_CREATE_NETIF_QP = 1 << 5, 647 /* reserve bits 26-31 for low level drivers' internal use */ 648 IB_QP_CREATE_RESERVED_START = 1 << 26, 649 IB_QP_CREATE_RESERVED_END = 1 << 31, 650 }; 651 652 653 /* 654 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler 655 * callback to destroy the passed in QP. 656 */ 657 658 struct ib_qp_init_attr { 659 void (*event_handler)(struct ib_event *, void *); 660 void *qp_context; 661 struct ib_cq *send_cq; 662 struct ib_cq *recv_cq; 663 struct ib_srq *srq; 664 struct ib_xrcd *xrcd; /* XRC TGT QPs only */ 665 struct ib_qp_cap cap; 666 enum ib_sig_type sq_sig_type; 667 enum ib_qp_type qp_type; 668 enum ib_qp_create_flags create_flags; 669 u8 port_num; /* special QP types only */ 670 }; 671 672 struct ib_qp_open_attr { 673 void (*event_handler)(struct ib_event *, void *); 674 void *qp_context; 675 u32 qp_num; 676 enum ib_qp_type qp_type; 677 }; 678 679 enum ib_rnr_timeout { 680 IB_RNR_TIMER_655_36 = 0, 681 IB_RNR_TIMER_000_01 = 1, 682 IB_RNR_TIMER_000_02 = 2, 683 IB_RNR_TIMER_000_03 = 3, 684 IB_RNR_TIMER_000_04 = 4, 685 IB_RNR_TIMER_000_06 = 5, 686 IB_RNR_TIMER_000_08 = 6, 687 IB_RNR_TIMER_000_12 = 7, 688 IB_RNR_TIMER_000_16 = 8, 689 IB_RNR_TIMER_000_24 = 9, 690 IB_RNR_TIMER_000_32 = 10, 691 IB_RNR_TIMER_000_48 = 11, 692 IB_RNR_TIMER_000_64 = 12, 693 IB_RNR_TIMER_000_96 = 13, 694 IB_RNR_TIMER_001_28 = 14, 695 IB_RNR_TIMER_001_92 = 15, 696 IB_RNR_TIMER_002_56 = 16, 697 IB_RNR_TIMER_003_84 = 17, 698 IB_RNR_TIMER_005_12 = 18, 699 IB_RNR_TIMER_007_68 = 19, 700 IB_RNR_TIMER_010_24 = 20, 701 IB_RNR_TIMER_015_36 = 21, 702 IB_RNR_TIMER_020_48 = 22, 703 IB_RNR_TIMER_030_72 = 23, 704 IB_RNR_TIMER_040_96 = 24, 705 IB_RNR_TIMER_061_44 = 25, 706 IB_RNR_TIMER_081_92 = 26, 707 IB_RNR_TIMER_122_88 = 27, 708 IB_RNR_TIMER_163_84 = 28, 709 IB_RNR_TIMER_245_76 = 29, 710 IB_RNR_TIMER_327_68 = 30, 711 IB_RNR_TIMER_491_52 = 31 712 }; 713 714 enum ib_qp_attr_mask { 715 IB_QP_STATE = 1, 716 IB_QP_CUR_STATE = (1<<1), 717 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2), 718 IB_QP_ACCESS_FLAGS = (1<<3), 719 IB_QP_PKEY_INDEX = (1<<4), 720 IB_QP_PORT = (1<<5), 721 IB_QP_QKEY = (1<<6), 722 IB_QP_AV = (1<<7), 723 IB_QP_PATH_MTU = (1<<8), 724 IB_QP_TIMEOUT = (1<<9), 725 IB_QP_RETRY_CNT = (1<<10), 726 IB_QP_RNR_RETRY = (1<<11), 727 IB_QP_RQ_PSN = (1<<12), 728 IB_QP_MAX_QP_RD_ATOMIC = (1<<13), 729 IB_QP_ALT_PATH = (1<<14), 730 IB_QP_MIN_RNR_TIMER = (1<<15), 731 IB_QP_SQ_PSN = (1<<16), 732 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17), 733 IB_QP_PATH_MIG_STATE = (1<<18), 734 IB_QP_CAP = (1<<19), 735 IB_QP_DEST_QPN = (1<<20), 736 IB_QP_SMAC = (1<<21), 737 IB_QP_ALT_SMAC = (1<<22), 738 IB_QP_VID = (1<<23), 739 IB_QP_ALT_VID = (1<<24), 740 }; 741 742 enum ib_qp_state { 743 IB_QPS_RESET, 744 IB_QPS_INIT, 745 IB_QPS_RTR, 746 IB_QPS_RTS, 747 IB_QPS_SQD, 748 IB_QPS_SQE, 749 IB_QPS_ERR 750 }; 751 752 enum ib_mig_state { 753 IB_MIG_MIGRATED, 754 IB_MIG_REARM, 755 IB_MIG_ARMED 756 }; 757 758 enum ib_mw_type { 759 IB_MW_TYPE_1 = 1, 760 IB_MW_TYPE_2 = 2 761 }; 762 763 struct ib_qp_attr { 764 enum ib_qp_state qp_state; 765 enum ib_qp_state cur_qp_state; 766 enum ib_mtu path_mtu; 767 enum ib_mig_state path_mig_state; 768 u32 qkey; 769 u32 rq_psn; 770 u32 sq_psn; 771 u32 dest_qp_num; 772 int qp_access_flags; 773 struct ib_qp_cap cap; 774 struct ib_ah_attr ah_attr; 775 struct ib_ah_attr alt_ah_attr; 776 u16 pkey_index; 777 u16 alt_pkey_index; 778 u8 en_sqd_async_notify; 779 u8 sq_draining; 780 u8 max_rd_atomic; 781 u8 max_dest_rd_atomic; 782 u8 min_rnr_timer; 783 u8 port_num; 784 u8 timeout; 785 u8 retry_cnt; 786 u8 rnr_retry; 787 u8 alt_port_num; 788 u8 alt_timeout; 789 u8 smac[ETH_ALEN]; 790 u8 alt_smac[ETH_ALEN]; 791 u16 vlan_id; 792 u16 alt_vlan_id; 793 }; 794 795 enum ib_wr_opcode { 796 IB_WR_RDMA_WRITE, 797 IB_WR_RDMA_WRITE_WITH_IMM, 798 IB_WR_SEND, 799 IB_WR_SEND_WITH_IMM, 800 IB_WR_RDMA_READ, 801 IB_WR_ATOMIC_CMP_AND_SWP, 802 IB_WR_ATOMIC_FETCH_AND_ADD, 803 IB_WR_LSO, 804 IB_WR_SEND_WITH_INV, 805 IB_WR_RDMA_READ_WITH_INV, 806 IB_WR_LOCAL_INV, 807 IB_WR_FAST_REG_MR, 808 IB_WR_MASKED_ATOMIC_CMP_AND_SWP, 809 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD, 810 IB_WR_BIND_MW, 811 /* reserve values for low level drivers' internal use. 812 * These values will not be used at all in the ib core layer. 813 */ 814 IB_WR_RESERVED1 = 0xf0, 815 IB_WR_RESERVED2, 816 IB_WR_RESERVED3, 817 IB_WR_RESERVED4, 818 IB_WR_RESERVED5, 819 IB_WR_RESERVED6, 820 IB_WR_RESERVED7, 821 IB_WR_RESERVED8, 822 IB_WR_RESERVED9, 823 IB_WR_RESERVED10, 824 }; 825 826 enum ib_send_flags { 827 IB_SEND_FENCE = 1, 828 IB_SEND_SIGNALED = (1<<1), 829 IB_SEND_SOLICITED = (1<<2), 830 IB_SEND_INLINE = (1<<3), 831 IB_SEND_IP_CSUM = (1<<4), 832 833 /* reserve bits 26-31 for low level drivers' internal use */ 834 IB_SEND_RESERVED_START = (1 << 26), 835 IB_SEND_RESERVED_END = (1 << 31), 836 }; 837 838 struct ib_sge { 839 u64 addr; 840 u32 length; 841 u32 lkey; 842 }; 843 844 struct ib_fast_reg_page_list { 845 struct ib_device *device; 846 u64 *page_list; 847 unsigned int max_page_list_len; 848 }; 849 850 /** 851 * struct ib_mw_bind_info - Parameters for a memory window bind operation. 852 * @mr: A memory region to bind the memory window to. 853 * @addr: The address where the memory window should begin. 854 * @length: The length of the memory window, in bytes. 855 * @mw_access_flags: Access flags from enum ib_access_flags for the window. 856 * 857 * This struct contains the shared parameters for type 1 and type 2 858 * memory window bind operations. 859 */ 860 struct ib_mw_bind_info { 861 struct ib_mr *mr; 862 u64 addr; 863 u64 length; 864 int mw_access_flags; 865 }; 866 867 struct ib_send_wr { 868 struct ib_send_wr *next; 869 u64 wr_id; 870 struct ib_sge *sg_list; 871 int num_sge; 872 enum ib_wr_opcode opcode; 873 int send_flags; 874 union { 875 __be32 imm_data; 876 u32 invalidate_rkey; 877 } ex; 878 union { 879 struct { 880 u64 remote_addr; 881 u32 rkey; 882 } rdma; 883 struct { 884 u64 remote_addr; 885 u64 compare_add; 886 u64 swap; 887 u64 compare_add_mask; 888 u64 swap_mask; 889 u32 rkey; 890 } atomic; 891 struct { 892 struct ib_ah *ah; 893 void *header; 894 int hlen; 895 int mss; 896 u32 remote_qpn; 897 u32 remote_qkey; 898 u16 pkey_index; /* valid for GSI only */ 899 u8 port_num; /* valid for DR SMPs on switch only */ 900 } ud; 901 struct { 902 u64 iova_start; 903 struct ib_fast_reg_page_list *page_list; 904 unsigned int page_shift; 905 unsigned int page_list_len; 906 u32 length; 907 int access_flags; 908 u32 rkey; 909 } fast_reg; 910 struct { 911 struct ib_mw *mw; 912 /* The new rkey for the memory window. */ 913 u32 rkey; 914 struct ib_mw_bind_info bind_info; 915 } bind_mw; 916 } wr; 917 u32 xrc_remote_srq_num; /* XRC TGT QPs only */ 918 }; 919 920 struct ib_recv_wr { 921 struct ib_recv_wr *next; 922 u64 wr_id; 923 struct ib_sge *sg_list; 924 int num_sge; 925 }; 926 927 enum ib_access_flags { 928 IB_ACCESS_LOCAL_WRITE = 1, 929 IB_ACCESS_REMOTE_WRITE = (1<<1), 930 IB_ACCESS_REMOTE_READ = (1<<2), 931 IB_ACCESS_REMOTE_ATOMIC = (1<<3), 932 IB_ACCESS_MW_BIND = (1<<4), 933 IB_ZERO_BASED = (1<<5) 934 }; 935 936 struct ib_phys_buf { 937 u64 addr; 938 u64 size; 939 }; 940 941 struct ib_mr_attr { 942 struct ib_pd *pd; 943 u64 device_virt_addr; 944 u64 size; 945 int mr_access_flags; 946 u32 lkey; 947 u32 rkey; 948 }; 949 950 enum ib_mr_rereg_flags { 951 IB_MR_REREG_TRANS = 1, 952 IB_MR_REREG_PD = (1<<1), 953 IB_MR_REREG_ACCESS = (1<<2) 954 }; 955 956 /** 957 * struct ib_mw_bind - Parameters for a type 1 memory window bind operation. 958 * @wr_id: Work request id. 959 * @send_flags: Flags from ib_send_flags enum. 960 * @bind_info: More parameters of the bind operation. 961 */ 962 struct ib_mw_bind { 963 u64 wr_id; 964 int send_flags; 965 struct ib_mw_bind_info bind_info; 966 }; 967 968 struct ib_fmr_attr { 969 int max_pages; 970 int max_maps; 971 u8 page_shift; 972 }; 973 974 struct ib_ucontext { 975 struct ib_device *device; 976 struct list_head pd_list; 977 struct list_head mr_list; 978 struct list_head mw_list; 979 struct list_head cq_list; 980 struct list_head qp_list; 981 struct list_head srq_list; 982 struct list_head ah_list; 983 struct list_head xrcd_list; 984 struct list_head rule_list; 985 int closing; 986 }; 987 988 struct ib_uobject { 989 u64 user_handle; /* handle given to us by userspace */ 990 struct ib_ucontext *context; /* associated user context */ 991 void *object; /* containing object */ 992 struct list_head list; /* link to context's list */ 993 int id; /* index into kernel idr */ 994 struct kref ref; 995 struct rw_semaphore mutex; /* protects .live */ 996 int live; 997 }; 998 999 struct ib_udata { 1000 const void __user *inbuf; 1001 void __user *outbuf; 1002 size_t inlen; 1003 size_t outlen; 1004 }; 1005 1006 struct ib_pd { 1007 struct ib_device *device; 1008 struct ib_uobject *uobject; 1009 atomic_t usecnt; /* count all resources */ 1010 }; 1011 1012 struct ib_xrcd { 1013 struct ib_device *device; 1014 atomic_t usecnt; /* count all exposed resources */ 1015 struct inode *inode; 1016 1017 struct mutex tgt_qp_mutex; 1018 struct list_head tgt_qp_list; 1019 }; 1020 1021 struct ib_ah { 1022 struct ib_device *device; 1023 struct ib_pd *pd; 1024 struct ib_uobject *uobject; 1025 }; 1026 1027 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context); 1028 1029 struct ib_cq { 1030 struct ib_device *device; 1031 struct ib_uobject *uobject; 1032 ib_comp_handler comp_handler; 1033 void (*event_handler)(struct ib_event *, void *); 1034 void *cq_context; 1035 int cqe; 1036 atomic_t usecnt; /* count number of work queues */ 1037 }; 1038 1039 struct ib_srq { 1040 struct ib_device *device; 1041 struct ib_pd *pd; 1042 struct ib_uobject *uobject; 1043 void (*event_handler)(struct ib_event *, void *); 1044 void *srq_context; 1045 enum ib_srq_type srq_type; 1046 atomic_t usecnt; 1047 1048 union { 1049 struct { 1050 struct ib_xrcd *xrcd; 1051 struct ib_cq *cq; 1052 u32 srq_num; 1053 } xrc; 1054 } ext; 1055 }; 1056 1057 struct ib_qp { 1058 struct ib_device *device; 1059 struct ib_pd *pd; 1060 struct ib_cq *send_cq; 1061 struct ib_cq *recv_cq; 1062 struct ib_srq *srq; 1063 struct ib_xrcd *xrcd; /* XRC TGT QPs only */ 1064 struct list_head xrcd_list; 1065 /* count times opened, mcast attaches, flow attaches */ 1066 atomic_t usecnt; 1067 struct list_head open_list; 1068 struct ib_qp *real_qp; 1069 struct ib_uobject *uobject; 1070 void (*event_handler)(struct ib_event *, void *); 1071 void *qp_context; 1072 u32 qp_num; 1073 enum ib_qp_type qp_type; 1074 }; 1075 1076 struct ib_mr { 1077 struct ib_device *device; 1078 struct ib_pd *pd; 1079 struct ib_uobject *uobject; 1080 u32 lkey; 1081 u32 rkey; 1082 atomic_t usecnt; /* count number of MWs */ 1083 }; 1084 1085 struct ib_mw { 1086 struct ib_device *device; 1087 struct ib_pd *pd; 1088 struct ib_uobject *uobject; 1089 u32 rkey; 1090 enum ib_mw_type type; 1091 }; 1092 1093 struct ib_fmr { 1094 struct ib_device *device; 1095 struct ib_pd *pd; 1096 struct list_head list; 1097 u32 lkey; 1098 u32 rkey; 1099 }; 1100 1101 /* Supported steering options */ 1102 enum ib_flow_attr_type { 1103 /* steering according to rule specifications */ 1104 IB_FLOW_ATTR_NORMAL = 0x0, 1105 /* default unicast and multicast rule - 1106 * receive all Eth traffic which isn't steered to any QP 1107 */ 1108 IB_FLOW_ATTR_ALL_DEFAULT = 0x1, 1109 /* default multicast rule - 1110 * receive all Eth multicast traffic which isn't steered to any QP 1111 */ 1112 IB_FLOW_ATTR_MC_DEFAULT = 0x2, 1113 /* sniffer rule - receive all port traffic */ 1114 IB_FLOW_ATTR_SNIFFER = 0x3 1115 }; 1116 1117 /* Supported steering header types */ 1118 enum ib_flow_spec_type { 1119 /* L2 headers*/ 1120 IB_FLOW_SPEC_ETH = 0x20, 1121 IB_FLOW_SPEC_IB = 0x22, 1122 /* L3 header*/ 1123 IB_FLOW_SPEC_IPV4 = 0x30, 1124 /* L4 headers*/ 1125 IB_FLOW_SPEC_TCP = 0x40, 1126 IB_FLOW_SPEC_UDP = 0x41 1127 }; 1128 #define IB_FLOW_SPEC_LAYER_MASK 0xF0 1129 #define IB_FLOW_SPEC_SUPPORT_LAYERS 4 1130 1131 /* Flow steering rule priority is set according to it's domain. 1132 * Lower domain value means higher priority. 1133 */ 1134 enum ib_flow_domain { 1135 IB_FLOW_DOMAIN_USER, 1136 IB_FLOW_DOMAIN_ETHTOOL, 1137 IB_FLOW_DOMAIN_RFS, 1138 IB_FLOW_DOMAIN_NIC, 1139 IB_FLOW_DOMAIN_NUM /* Must be last */ 1140 }; 1141 1142 struct ib_flow_eth_filter { 1143 u8 dst_mac[6]; 1144 u8 src_mac[6]; 1145 __be16 ether_type; 1146 __be16 vlan_tag; 1147 }; 1148 1149 struct ib_flow_spec_eth { 1150 enum ib_flow_spec_type type; 1151 u16 size; 1152 struct ib_flow_eth_filter val; 1153 struct ib_flow_eth_filter mask; 1154 }; 1155 1156 struct ib_flow_ib_filter { 1157 __be16 dlid; 1158 __u8 sl; 1159 }; 1160 1161 struct ib_flow_spec_ib { 1162 enum ib_flow_spec_type type; 1163 u16 size; 1164 struct ib_flow_ib_filter val; 1165 struct ib_flow_ib_filter mask; 1166 }; 1167 1168 struct ib_flow_ipv4_filter { 1169 __be32 src_ip; 1170 __be32 dst_ip; 1171 }; 1172 1173 struct ib_flow_spec_ipv4 { 1174 enum ib_flow_spec_type type; 1175 u16 size; 1176 struct ib_flow_ipv4_filter val; 1177 struct ib_flow_ipv4_filter mask; 1178 }; 1179 1180 struct ib_flow_tcp_udp_filter { 1181 __be16 dst_port; 1182 __be16 src_port; 1183 }; 1184 1185 struct ib_flow_spec_tcp_udp { 1186 enum ib_flow_spec_type type; 1187 u16 size; 1188 struct ib_flow_tcp_udp_filter val; 1189 struct ib_flow_tcp_udp_filter mask; 1190 }; 1191 1192 union ib_flow_spec { 1193 struct { 1194 enum ib_flow_spec_type type; 1195 u16 size; 1196 }; 1197 struct ib_flow_spec_eth eth; 1198 struct ib_flow_spec_ib ib; 1199 struct ib_flow_spec_ipv4 ipv4; 1200 struct ib_flow_spec_tcp_udp tcp_udp; 1201 }; 1202 1203 struct ib_flow_attr { 1204 enum ib_flow_attr_type type; 1205 u16 size; 1206 u16 priority; 1207 u32 flags; 1208 u8 num_of_specs; 1209 u8 port; 1210 /* Following are the optional layers according to user request 1211 * struct ib_flow_spec_xxx 1212 * struct ib_flow_spec_yyy 1213 */ 1214 }; 1215 1216 struct ib_flow { 1217 struct ib_qp *qp; 1218 struct ib_uobject *uobject; 1219 }; 1220 1221 struct ib_mad; 1222 struct ib_grh; 1223 1224 enum ib_process_mad_flags { 1225 IB_MAD_IGNORE_MKEY = 1, 1226 IB_MAD_IGNORE_BKEY = 2, 1227 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY 1228 }; 1229 1230 enum ib_mad_result { 1231 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */ 1232 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */ 1233 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */ 1234 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */ 1235 }; 1236 1237 #define IB_DEVICE_NAME_MAX 64 1238 1239 struct ib_cache { 1240 rwlock_t lock; 1241 struct ib_event_handler event_handler; 1242 struct ib_pkey_cache **pkey_cache; 1243 struct ib_gid_cache **gid_cache; 1244 u8 *lmc_cache; 1245 }; 1246 1247 struct ib_dma_mapping_ops { 1248 int (*mapping_error)(struct ib_device *dev, 1249 u64 dma_addr); 1250 u64 (*map_single)(struct ib_device *dev, 1251 void *ptr, size_t size, 1252 enum dma_data_direction direction); 1253 void (*unmap_single)(struct ib_device *dev, 1254 u64 addr, size_t size, 1255 enum dma_data_direction direction); 1256 u64 (*map_page)(struct ib_device *dev, 1257 struct page *page, unsigned long offset, 1258 size_t size, 1259 enum dma_data_direction direction); 1260 void (*unmap_page)(struct ib_device *dev, 1261 u64 addr, size_t size, 1262 enum dma_data_direction direction); 1263 int (*map_sg)(struct ib_device *dev, 1264 struct scatterlist *sg, int nents, 1265 enum dma_data_direction direction); 1266 void (*unmap_sg)(struct ib_device *dev, 1267 struct scatterlist *sg, int nents, 1268 enum dma_data_direction direction); 1269 u64 (*dma_address)(struct ib_device *dev, 1270 struct scatterlist *sg); 1271 unsigned int (*dma_len)(struct ib_device *dev, 1272 struct scatterlist *sg); 1273 void (*sync_single_for_cpu)(struct ib_device *dev, 1274 u64 dma_handle, 1275 size_t size, 1276 enum dma_data_direction dir); 1277 void (*sync_single_for_device)(struct ib_device *dev, 1278 u64 dma_handle, 1279 size_t size, 1280 enum dma_data_direction dir); 1281 void *(*alloc_coherent)(struct ib_device *dev, 1282 size_t size, 1283 u64 *dma_handle, 1284 gfp_t flag); 1285 void (*free_coherent)(struct ib_device *dev, 1286 size_t size, void *cpu_addr, 1287 u64 dma_handle); 1288 }; 1289 1290 struct iw_cm_verbs; 1291 1292 struct ib_device { 1293 struct device *dma_device; 1294 1295 char name[IB_DEVICE_NAME_MAX]; 1296 1297 struct list_head event_handler_list; 1298 spinlock_t event_handler_lock; 1299 1300 spinlock_t client_data_lock; 1301 struct list_head core_list; 1302 struct list_head client_data_list; 1303 1304 struct ib_cache cache; 1305 int *pkey_tbl_len; 1306 int *gid_tbl_len; 1307 1308 int num_comp_vectors; 1309 1310 struct iw_cm_verbs *iwcm; 1311 1312 int (*get_protocol_stats)(struct ib_device *device, 1313 union rdma_protocol_stats *stats); 1314 int (*query_device)(struct ib_device *device, 1315 struct ib_device_attr *device_attr); 1316 int (*query_port)(struct ib_device *device, 1317 u8 port_num, 1318 struct ib_port_attr *port_attr); 1319 enum rdma_link_layer (*get_link_layer)(struct ib_device *device, 1320 u8 port_num); 1321 int (*query_gid)(struct ib_device *device, 1322 u8 port_num, int index, 1323 union ib_gid *gid); 1324 int (*query_pkey)(struct ib_device *device, 1325 u8 port_num, u16 index, u16 *pkey); 1326 int (*modify_device)(struct ib_device *device, 1327 int device_modify_mask, 1328 struct ib_device_modify *device_modify); 1329 int (*modify_port)(struct ib_device *device, 1330 u8 port_num, int port_modify_mask, 1331 struct ib_port_modify *port_modify); 1332 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device, 1333 struct ib_udata *udata); 1334 int (*dealloc_ucontext)(struct ib_ucontext *context); 1335 int (*mmap)(struct ib_ucontext *context, 1336 struct vm_area_struct *vma); 1337 struct ib_pd * (*alloc_pd)(struct ib_device *device, 1338 struct ib_ucontext *context, 1339 struct ib_udata *udata); 1340 int (*dealloc_pd)(struct ib_pd *pd); 1341 struct ib_ah * (*create_ah)(struct ib_pd *pd, 1342 struct ib_ah_attr *ah_attr); 1343 int (*modify_ah)(struct ib_ah *ah, 1344 struct ib_ah_attr *ah_attr); 1345 int (*query_ah)(struct ib_ah *ah, 1346 struct ib_ah_attr *ah_attr); 1347 int (*destroy_ah)(struct ib_ah *ah); 1348 struct ib_srq * (*create_srq)(struct ib_pd *pd, 1349 struct ib_srq_init_attr *srq_init_attr, 1350 struct ib_udata *udata); 1351 int (*modify_srq)(struct ib_srq *srq, 1352 struct ib_srq_attr *srq_attr, 1353 enum ib_srq_attr_mask srq_attr_mask, 1354 struct ib_udata *udata); 1355 int (*query_srq)(struct ib_srq *srq, 1356 struct ib_srq_attr *srq_attr); 1357 int (*destroy_srq)(struct ib_srq *srq); 1358 int (*post_srq_recv)(struct ib_srq *srq, 1359 struct ib_recv_wr *recv_wr, 1360 struct ib_recv_wr **bad_recv_wr); 1361 struct ib_qp * (*create_qp)(struct ib_pd *pd, 1362 struct ib_qp_init_attr *qp_init_attr, 1363 struct ib_udata *udata); 1364 int (*modify_qp)(struct ib_qp *qp, 1365 struct ib_qp_attr *qp_attr, 1366 int qp_attr_mask, 1367 struct ib_udata *udata); 1368 int (*query_qp)(struct ib_qp *qp, 1369 struct ib_qp_attr *qp_attr, 1370 int qp_attr_mask, 1371 struct ib_qp_init_attr *qp_init_attr); 1372 int (*destroy_qp)(struct ib_qp *qp); 1373 int (*post_send)(struct ib_qp *qp, 1374 struct ib_send_wr *send_wr, 1375 struct ib_send_wr **bad_send_wr); 1376 int (*post_recv)(struct ib_qp *qp, 1377 struct ib_recv_wr *recv_wr, 1378 struct ib_recv_wr **bad_recv_wr); 1379 struct ib_cq * (*create_cq)(struct ib_device *device, int cqe, 1380 int comp_vector, 1381 struct ib_ucontext *context, 1382 struct ib_udata *udata); 1383 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, 1384 u16 cq_period); 1385 int (*destroy_cq)(struct ib_cq *cq); 1386 int (*resize_cq)(struct ib_cq *cq, int cqe, 1387 struct ib_udata *udata); 1388 int (*poll_cq)(struct ib_cq *cq, int num_entries, 1389 struct ib_wc *wc); 1390 int (*peek_cq)(struct ib_cq *cq, int wc_cnt); 1391 int (*req_notify_cq)(struct ib_cq *cq, 1392 enum ib_cq_notify_flags flags); 1393 int (*req_ncomp_notif)(struct ib_cq *cq, 1394 int wc_cnt); 1395 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd, 1396 int mr_access_flags); 1397 struct ib_mr * (*reg_phys_mr)(struct ib_pd *pd, 1398 struct ib_phys_buf *phys_buf_array, 1399 int num_phys_buf, 1400 int mr_access_flags, 1401 u64 *iova_start); 1402 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd, 1403 u64 start, u64 length, 1404 u64 virt_addr, 1405 int mr_access_flags, 1406 struct ib_udata *udata); 1407 int (*query_mr)(struct ib_mr *mr, 1408 struct ib_mr_attr *mr_attr); 1409 int (*dereg_mr)(struct ib_mr *mr); 1410 struct ib_mr * (*alloc_fast_reg_mr)(struct ib_pd *pd, 1411 int max_page_list_len); 1412 struct ib_fast_reg_page_list * (*alloc_fast_reg_page_list)(struct ib_device *device, 1413 int page_list_len); 1414 void (*free_fast_reg_page_list)(struct ib_fast_reg_page_list *page_list); 1415 int (*rereg_phys_mr)(struct ib_mr *mr, 1416 int mr_rereg_mask, 1417 struct ib_pd *pd, 1418 struct ib_phys_buf *phys_buf_array, 1419 int num_phys_buf, 1420 int mr_access_flags, 1421 u64 *iova_start); 1422 struct ib_mw * (*alloc_mw)(struct ib_pd *pd, 1423 enum ib_mw_type type); 1424 int (*bind_mw)(struct ib_qp *qp, 1425 struct ib_mw *mw, 1426 struct ib_mw_bind *mw_bind); 1427 int (*dealloc_mw)(struct ib_mw *mw); 1428 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd, 1429 int mr_access_flags, 1430 struct ib_fmr_attr *fmr_attr); 1431 int (*map_phys_fmr)(struct ib_fmr *fmr, 1432 u64 *page_list, int list_len, 1433 u64 iova); 1434 int (*unmap_fmr)(struct list_head *fmr_list); 1435 int (*dealloc_fmr)(struct ib_fmr *fmr); 1436 int (*attach_mcast)(struct ib_qp *qp, 1437 union ib_gid *gid, 1438 u16 lid); 1439 int (*detach_mcast)(struct ib_qp *qp, 1440 union ib_gid *gid, 1441 u16 lid); 1442 int (*process_mad)(struct ib_device *device, 1443 int process_mad_flags, 1444 u8 port_num, 1445 struct ib_wc *in_wc, 1446 struct ib_grh *in_grh, 1447 struct ib_mad *in_mad, 1448 struct ib_mad *out_mad); 1449 struct ib_xrcd * (*alloc_xrcd)(struct ib_device *device, 1450 struct ib_ucontext *ucontext, 1451 struct ib_udata *udata); 1452 int (*dealloc_xrcd)(struct ib_xrcd *xrcd); 1453 struct ib_flow * (*create_flow)(struct ib_qp *qp, 1454 struct ib_flow_attr 1455 *flow_attr, 1456 int domain); 1457 int (*destroy_flow)(struct ib_flow *flow_id); 1458 1459 struct ib_dma_mapping_ops *dma_ops; 1460 1461 struct module *owner; 1462 struct device dev; 1463 struct kobject *ports_parent; 1464 struct list_head port_list; 1465 1466 enum { 1467 IB_DEV_UNINITIALIZED, 1468 IB_DEV_REGISTERED, 1469 IB_DEV_UNREGISTERED 1470 } reg_state; 1471 1472 int uverbs_abi_ver; 1473 u64 uverbs_cmd_mask; 1474 u64 uverbs_ex_cmd_mask; 1475 1476 char node_desc[64]; 1477 __be64 node_guid; 1478 u32 local_dma_lkey; 1479 u8 node_type; 1480 u8 phys_port_cnt; 1481 }; 1482 1483 struct ib_client { 1484 char *name; 1485 void (*add) (struct ib_device *); 1486 void (*remove)(struct ib_device *); 1487 1488 struct list_head list; 1489 }; 1490 1491 struct ib_device *ib_alloc_device(size_t size); 1492 void ib_dealloc_device(struct ib_device *device); 1493 1494 int ib_register_device(struct ib_device *device, 1495 int (*port_callback)(struct ib_device *, 1496 u8, struct kobject *)); 1497 void ib_unregister_device(struct ib_device *device); 1498 1499 int ib_register_client (struct ib_client *client); 1500 void ib_unregister_client(struct ib_client *client); 1501 1502 void *ib_get_client_data(struct ib_device *device, struct ib_client *client); 1503 void ib_set_client_data(struct ib_device *device, struct ib_client *client, 1504 void *data); 1505 1506 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len) 1507 { 1508 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0; 1509 } 1510 1511 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len) 1512 { 1513 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0; 1514 } 1515 1516 /** 1517 * ib_modify_qp_is_ok - Check that the supplied attribute mask 1518 * contains all required attributes and no attributes not allowed for 1519 * the given QP state transition. 1520 * @cur_state: Current QP state 1521 * @next_state: Next QP state 1522 * @type: QP type 1523 * @mask: Mask of supplied QP attributes 1524 * @ll : link layer of port 1525 * 1526 * This function is a helper function that a low-level driver's 1527 * modify_qp method can use to validate the consumer's input. It 1528 * checks that cur_state and next_state are valid QP states, that a 1529 * transition from cur_state to next_state is allowed by the IB spec, 1530 * and that the attribute mask supplied is allowed for the transition. 1531 */ 1532 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state, 1533 enum ib_qp_type type, enum ib_qp_attr_mask mask, 1534 enum rdma_link_layer ll); 1535 1536 int ib_register_event_handler (struct ib_event_handler *event_handler); 1537 int ib_unregister_event_handler(struct ib_event_handler *event_handler); 1538 void ib_dispatch_event(struct ib_event *event); 1539 1540 int ib_query_device(struct ib_device *device, 1541 struct ib_device_attr *device_attr); 1542 1543 int ib_query_port(struct ib_device *device, 1544 u8 port_num, struct ib_port_attr *port_attr); 1545 1546 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device, 1547 u8 port_num); 1548 1549 int ib_query_gid(struct ib_device *device, 1550 u8 port_num, int index, union ib_gid *gid); 1551 1552 int ib_query_pkey(struct ib_device *device, 1553 u8 port_num, u16 index, u16 *pkey); 1554 1555 int ib_modify_device(struct ib_device *device, 1556 int device_modify_mask, 1557 struct ib_device_modify *device_modify); 1558 1559 int ib_modify_port(struct ib_device *device, 1560 u8 port_num, int port_modify_mask, 1561 struct ib_port_modify *port_modify); 1562 1563 int ib_find_gid(struct ib_device *device, union ib_gid *gid, 1564 u8 *port_num, u16 *index); 1565 1566 int ib_find_pkey(struct ib_device *device, 1567 u8 port_num, u16 pkey, u16 *index); 1568 1569 /** 1570 * ib_alloc_pd - Allocates an unused protection domain. 1571 * @device: The device on which to allocate the protection domain. 1572 * 1573 * A protection domain object provides an association between QPs, shared 1574 * receive queues, address handles, memory regions, and memory windows. 1575 */ 1576 struct ib_pd *ib_alloc_pd(struct ib_device *device); 1577 1578 /** 1579 * ib_dealloc_pd - Deallocates a protection domain. 1580 * @pd: The protection domain to deallocate. 1581 */ 1582 int ib_dealloc_pd(struct ib_pd *pd); 1583 1584 /** 1585 * ib_create_ah - Creates an address handle for the given address vector. 1586 * @pd: The protection domain associated with the address handle. 1587 * @ah_attr: The attributes of the address vector. 1588 * 1589 * The address handle is used to reference a local or global destination 1590 * in all UD QP post sends. 1591 */ 1592 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr); 1593 1594 /** 1595 * ib_init_ah_from_wc - Initializes address handle attributes from a 1596 * work completion. 1597 * @device: Device on which the received message arrived. 1598 * @port_num: Port on which the received message arrived. 1599 * @wc: Work completion associated with the received message. 1600 * @grh: References the received global route header. This parameter is 1601 * ignored unless the work completion indicates that the GRH is valid. 1602 * @ah_attr: Returned attributes that can be used when creating an address 1603 * handle for replying to the message. 1604 */ 1605 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num, struct ib_wc *wc, 1606 struct ib_grh *grh, struct ib_ah_attr *ah_attr); 1607 1608 /** 1609 * ib_create_ah_from_wc - Creates an address handle associated with the 1610 * sender of the specified work completion. 1611 * @pd: The protection domain associated with the address handle. 1612 * @wc: Work completion information associated with a received message. 1613 * @grh: References the received global route header. This parameter is 1614 * ignored unless the work completion indicates that the GRH is valid. 1615 * @port_num: The outbound port number to associate with the address. 1616 * 1617 * The address handle is used to reference a local or global destination 1618 * in all UD QP post sends. 1619 */ 1620 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, struct ib_wc *wc, 1621 struct ib_grh *grh, u8 port_num); 1622 1623 /** 1624 * ib_modify_ah - Modifies the address vector associated with an address 1625 * handle. 1626 * @ah: The address handle to modify. 1627 * @ah_attr: The new address vector attributes to associate with the 1628 * address handle. 1629 */ 1630 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr); 1631 1632 /** 1633 * ib_query_ah - Queries the address vector associated with an address 1634 * handle. 1635 * @ah: The address handle to query. 1636 * @ah_attr: The address vector attributes associated with the address 1637 * handle. 1638 */ 1639 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr); 1640 1641 /** 1642 * ib_destroy_ah - Destroys an address handle. 1643 * @ah: The address handle to destroy. 1644 */ 1645 int ib_destroy_ah(struct ib_ah *ah); 1646 1647 /** 1648 * ib_create_srq - Creates a SRQ associated with the specified protection 1649 * domain. 1650 * @pd: The protection domain associated with the SRQ. 1651 * @srq_init_attr: A list of initial attributes required to create the 1652 * SRQ. If SRQ creation succeeds, then the attributes are updated to 1653 * the actual capabilities of the created SRQ. 1654 * 1655 * srq_attr->max_wr and srq_attr->max_sge are read the determine the 1656 * requested size of the SRQ, and set to the actual values allocated 1657 * on return. If ib_create_srq() succeeds, then max_wr and max_sge 1658 * will always be at least as large as the requested values. 1659 */ 1660 struct ib_srq *ib_create_srq(struct ib_pd *pd, 1661 struct ib_srq_init_attr *srq_init_attr); 1662 1663 /** 1664 * ib_modify_srq - Modifies the attributes for the specified SRQ. 1665 * @srq: The SRQ to modify. 1666 * @srq_attr: On input, specifies the SRQ attributes to modify. On output, 1667 * the current values of selected SRQ attributes are returned. 1668 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ 1669 * are being modified. 1670 * 1671 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or 1672 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when 1673 * the number of receives queued drops below the limit. 1674 */ 1675 int ib_modify_srq(struct ib_srq *srq, 1676 struct ib_srq_attr *srq_attr, 1677 enum ib_srq_attr_mask srq_attr_mask); 1678 1679 /** 1680 * ib_query_srq - Returns the attribute list and current values for the 1681 * specified SRQ. 1682 * @srq: The SRQ to query. 1683 * @srq_attr: The attributes of the specified SRQ. 1684 */ 1685 int ib_query_srq(struct ib_srq *srq, 1686 struct ib_srq_attr *srq_attr); 1687 1688 /** 1689 * ib_destroy_srq - Destroys the specified SRQ. 1690 * @srq: The SRQ to destroy. 1691 */ 1692 int ib_destroy_srq(struct ib_srq *srq); 1693 1694 /** 1695 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ. 1696 * @srq: The SRQ to post the work request on. 1697 * @recv_wr: A list of work requests to post on the receive queue. 1698 * @bad_recv_wr: On an immediate failure, this parameter will reference 1699 * the work request that failed to be posted on the QP. 1700 */ 1701 static inline int ib_post_srq_recv(struct ib_srq *srq, 1702 struct ib_recv_wr *recv_wr, 1703 struct ib_recv_wr **bad_recv_wr) 1704 { 1705 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr); 1706 } 1707 1708 /** 1709 * ib_create_qp - Creates a QP associated with the specified protection 1710 * domain. 1711 * @pd: The protection domain associated with the QP. 1712 * @qp_init_attr: A list of initial attributes required to create the 1713 * QP. If QP creation succeeds, then the attributes are updated to 1714 * the actual capabilities of the created QP. 1715 */ 1716 struct ib_qp *ib_create_qp(struct ib_pd *pd, 1717 struct ib_qp_init_attr *qp_init_attr); 1718 1719 /** 1720 * ib_modify_qp - Modifies the attributes for the specified QP and then 1721 * transitions the QP to the given state. 1722 * @qp: The QP to modify. 1723 * @qp_attr: On input, specifies the QP attributes to modify. On output, 1724 * the current values of selected QP attributes are returned. 1725 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP 1726 * are being modified. 1727 */ 1728 int ib_modify_qp(struct ib_qp *qp, 1729 struct ib_qp_attr *qp_attr, 1730 int qp_attr_mask); 1731 1732 /** 1733 * ib_query_qp - Returns the attribute list and current values for the 1734 * specified QP. 1735 * @qp: The QP to query. 1736 * @qp_attr: The attributes of the specified QP. 1737 * @qp_attr_mask: A bit-mask used to select specific attributes to query. 1738 * @qp_init_attr: Additional attributes of the selected QP. 1739 * 1740 * The qp_attr_mask may be used to limit the query to gathering only the 1741 * selected attributes. 1742 */ 1743 int ib_query_qp(struct ib_qp *qp, 1744 struct ib_qp_attr *qp_attr, 1745 int qp_attr_mask, 1746 struct ib_qp_init_attr *qp_init_attr); 1747 1748 /** 1749 * ib_destroy_qp - Destroys the specified QP. 1750 * @qp: The QP to destroy. 1751 */ 1752 int ib_destroy_qp(struct ib_qp *qp); 1753 1754 /** 1755 * ib_open_qp - Obtain a reference to an existing sharable QP. 1756 * @xrcd - XRC domain 1757 * @qp_open_attr: Attributes identifying the QP to open. 1758 * 1759 * Returns a reference to a sharable QP. 1760 */ 1761 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd, 1762 struct ib_qp_open_attr *qp_open_attr); 1763 1764 /** 1765 * ib_close_qp - Release an external reference to a QP. 1766 * @qp: The QP handle to release 1767 * 1768 * The opened QP handle is released by the caller. The underlying 1769 * shared QP is not destroyed until all internal references are released. 1770 */ 1771 int ib_close_qp(struct ib_qp *qp); 1772 1773 /** 1774 * ib_post_send - Posts a list of work requests to the send queue of 1775 * the specified QP. 1776 * @qp: The QP to post the work request on. 1777 * @send_wr: A list of work requests to post on the send queue. 1778 * @bad_send_wr: On an immediate failure, this parameter will reference 1779 * the work request that failed to be posted on the QP. 1780 * 1781 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate 1782 * error is returned, the QP state shall not be affected, 1783 * ib_post_send() will return an immediate error after queueing any 1784 * earlier work requests in the list. 1785 */ 1786 static inline int ib_post_send(struct ib_qp *qp, 1787 struct ib_send_wr *send_wr, 1788 struct ib_send_wr **bad_send_wr) 1789 { 1790 return qp->device->post_send(qp, send_wr, bad_send_wr); 1791 } 1792 1793 /** 1794 * ib_post_recv - Posts a list of work requests to the receive queue of 1795 * the specified QP. 1796 * @qp: The QP to post the work request on. 1797 * @recv_wr: A list of work requests to post on the receive queue. 1798 * @bad_recv_wr: On an immediate failure, this parameter will reference 1799 * the work request that failed to be posted on the QP. 1800 */ 1801 static inline int ib_post_recv(struct ib_qp *qp, 1802 struct ib_recv_wr *recv_wr, 1803 struct ib_recv_wr **bad_recv_wr) 1804 { 1805 return qp->device->post_recv(qp, recv_wr, bad_recv_wr); 1806 } 1807 1808 /** 1809 * ib_create_cq - Creates a CQ on the specified device. 1810 * @device: The device on which to create the CQ. 1811 * @comp_handler: A user-specified callback that is invoked when a 1812 * completion event occurs on the CQ. 1813 * @event_handler: A user-specified callback that is invoked when an 1814 * asynchronous event not associated with a completion occurs on the CQ. 1815 * @cq_context: Context associated with the CQ returned to the user via 1816 * the associated completion and event handlers. 1817 * @cqe: The minimum size of the CQ. 1818 * @comp_vector - Completion vector used to signal completion events. 1819 * Must be >= 0 and < context->num_comp_vectors. 1820 * 1821 * Users can examine the cq structure to determine the actual CQ size. 1822 */ 1823 struct ib_cq *ib_create_cq(struct ib_device *device, 1824 ib_comp_handler comp_handler, 1825 void (*event_handler)(struct ib_event *, void *), 1826 void *cq_context, int cqe, int comp_vector); 1827 1828 /** 1829 * ib_resize_cq - Modifies the capacity of the CQ. 1830 * @cq: The CQ to resize. 1831 * @cqe: The minimum size of the CQ. 1832 * 1833 * Users can examine the cq structure to determine the actual CQ size. 1834 */ 1835 int ib_resize_cq(struct ib_cq *cq, int cqe); 1836 1837 /** 1838 * ib_modify_cq - Modifies moderation params of the CQ 1839 * @cq: The CQ to modify. 1840 * @cq_count: number of CQEs that will trigger an event 1841 * @cq_period: max period of time in usec before triggering an event 1842 * 1843 */ 1844 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period); 1845 1846 /** 1847 * ib_destroy_cq - Destroys the specified CQ. 1848 * @cq: The CQ to destroy. 1849 */ 1850 int ib_destroy_cq(struct ib_cq *cq); 1851 1852 /** 1853 * ib_poll_cq - poll a CQ for completion(s) 1854 * @cq:the CQ being polled 1855 * @num_entries:maximum number of completions to return 1856 * @wc:array of at least @num_entries &struct ib_wc where completions 1857 * will be returned 1858 * 1859 * Poll a CQ for (possibly multiple) completions. If the return value 1860 * is < 0, an error occurred. If the return value is >= 0, it is the 1861 * number of completions returned. If the return value is 1862 * non-negative and < num_entries, then the CQ was emptied. 1863 */ 1864 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries, 1865 struct ib_wc *wc) 1866 { 1867 return cq->device->poll_cq(cq, num_entries, wc); 1868 } 1869 1870 /** 1871 * ib_peek_cq - Returns the number of unreaped completions currently 1872 * on the specified CQ. 1873 * @cq: The CQ to peek. 1874 * @wc_cnt: A minimum number of unreaped completions to check for. 1875 * 1876 * If the number of unreaped completions is greater than or equal to wc_cnt, 1877 * this function returns wc_cnt, otherwise, it returns the actual number of 1878 * unreaped completions. 1879 */ 1880 int ib_peek_cq(struct ib_cq *cq, int wc_cnt); 1881 1882 /** 1883 * ib_req_notify_cq - Request completion notification on a CQ. 1884 * @cq: The CQ to generate an event for. 1885 * @flags: 1886 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP 1887 * to request an event on the next solicited event or next work 1888 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS 1889 * may also be |ed in to request a hint about missed events, as 1890 * described below. 1891 * 1892 * Return Value: 1893 * < 0 means an error occurred while requesting notification 1894 * == 0 means notification was requested successfully, and if 1895 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events 1896 * were missed and it is safe to wait for another event. In 1897 * this case is it guaranteed that any work completions added 1898 * to the CQ since the last CQ poll will trigger a completion 1899 * notification event. 1900 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed 1901 * in. It means that the consumer must poll the CQ again to 1902 * make sure it is empty to avoid missing an event because of a 1903 * race between requesting notification and an entry being 1904 * added to the CQ. This return value means it is possible 1905 * (but not guaranteed) that a work completion has been added 1906 * to the CQ since the last poll without triggering a 1907 * completion notification event. 1908 */ 1909 static inline int ib_req_notify_cq(struct ib_cq *cq, 1910 enum ib_cq_notify_flags flags) 1911 { 1912 return cq->device->req_notify_cq(cq, flags); 1913 } 1914 1915 /** 1916 * ib_req_ncomp_notif - Request completion notification when there are 1917 * at least the specified number of unreaped completions on the CQ. 1918 * @cq: The CQ to generate an event for. 1919 * @wc_cnt: The number of unreaped completions that should be on the 1920 * CQ before an event is generated. 1921 */ 1922 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt) 1923 { 1924 return cq->device->req_ncomp_notif ? 1925 cq->device->req_ncomp_notif(cq, wc_cnt) : 1926 -ENOSYS; 1927 } 1928 1929 /** 1930 * ib_get_dma_mr - Returns a memory region for system memory that is 1931 * usable for DMA. 1932 * @pd: The protection domain associated with the memory region. 1933 * @mr_access_flags: Specifies the memory access rights. 1934 * 1935 * Note that the ib_dma_*() functions defined below must be used 1936 * to create/destroy addresses used with the Lkey or Rkey returned 1937 * by ib_get_dma_mr(). 1938 */ 1939 struct ib_mr *ib_get_dma_mr(struct ib_pd *pd, int mr_access_flags); 1940 1941 /** 1942 * ib_dma_mapping_error - check a DMA addr for error 1943 * @dev: The device for which the dma_addr was created 1944 * @dma_addr: The DMA address to check 1945 */ 1946 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr) 1947 { 1948 if (dev->dma_ops) 1949 return dev->dma_ops->mapping_error(dev, dma_addr); 1950 return dma_mapping_error(dev->dma_device, dma_addr); 1951 } 1952 1953 /** 1954 * ib_dma_map_single - Map a kernel virtual address to DMA address 1955 * @dev: The device for which the dma_addr is to be created 1956 * @cpu_addr: The kernel virtual address 1957 * @size: The size of the region in bytes 1958 * @direction: The direction of the DMA 1959 */ 1960 static inline u64 ib_dma_map_single(struct ib_device *dev, 1961 void *cpu_addr, size_t size, 1962 enum dma_data_direction direction) 1963 { 1964 if (dev->dma_ops) 1965 return dev->dma_ops->map_single(dev, cpu_addr, size, direction); 1966 return dma_map_single(dev->dma_device, cpu_addr, size, direction); 1967 } 1968 1969 /** 1970 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single() 1971 * @dev: The device for which the DMA address was created 1972 * @addr: The DMA address 1973 * @size: The size of the region in bytes 1974 * @direction: The direction of the DMA 1975 */ 1976 static inline void ib_dma_unmap_single(struct ib_device *dev, 1977 u64 addr, size_t size, 1978 enum dma_data_direction direction) 1979 { 1980 if (dev->dma_ops) 1981 dev->dma_ops->unmap_single(dev, addr, size, direction); 1982 else 1983 dma_unmap_single(dev->dma_device, addr, size, direction); 1984 } 1985 1986 static inline u64 ib_dma_map_single_attrs(struct ib_device *dev, 1987 void *cpu_addr, size_t size, 1988 enum dma_data_direction direction, 1989 struct dma_attrs *attrs) 1990 { 1991 return dma_map_single_attrs(dev->dma_device, cpu_addr, size, 1992 direction, attrs); 1993 } 1994 1995 static inline void ib_dma_unmap_single_attrs(struct ib_device *dev, 1996 u64 addr, size_t size, 1997 enum dma_data_direction direction, 1998 struct dma_attrs *attrs) 1999 { 2000 return dma_unmap_single_attrs(dev->dma_device, addr, size, 2001 direction, attrs); 2002 } 2003 2004 /** 2005 * ib_dma_map_page - Map a physical page to DMA address 2006 * @dev: The device for which the dma_addr is to be created 2007 * @page: The page to be mapped 2008 * @offset: The offset within the page 2009 * @size: The size of the region in bytes 2010 * @direction: The direction of the DMA 2011 */ 2012 static inline u64 ib_dma_map_page(struct ib_device *dev, 2013 struct page *page, 2014 unsigned long offset, 2015 size_t size, 2016 enum dma_data_direction direction) 2017 { 2018 if (dev->dma_ops) 2019 return dev->dma_ops->map_page(dev, page, offset, size, direction); 2020 return dma_map_page(dev->dma_device, page, offset, size, direction); 2021 } 2022 2023 /** 2024 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page() 2025 * @dev: The device for which the DMA address was created 2026 * @addr: The DMA address 2027 * @size: The size of the region in bytes 2028 * @direction: The direction of the DMA 2029 */ 2030 static inline void ib_dma_unmap_page(struct ib_device *dev, 2031 u64 addr, size_t size, 2032 enum dma_data_direction direction) 2033 { 2034 if (dev->dma_ops) 2035 dev->dma_ops->unmap_page(dev, addr, size, direction); 2036 else 2037 dma_unmap_page(dev->dma_device, addr, size, direction); 2038 } 2039 2040 /** 2041 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses 2042 * @dev: The device for which the DMA addresses are to be created 2043 * @sg: The array of scatter/gather entries 2044 * @nents: The number of scatter/gather entries 2045 * @direction: The direction of the DMA 2046 */ 2047 static inline int ib_dma_map_sg(struct ib_device *dev, 2048 struct scatterlist *sg, int nents, 2049 enum dma_data_direction direction) 2050 { 2051 if (dev->dma_ops) 2052 return dev->dma_ops->map_sg(dev, sg, nents, direction); 2053 return dma_map_sg(dev->dma_device, sg, nents, direction); 2054 } 2055 2056 /** 2057 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses 2058 * @dev: The device for which the DMA addresses were created 2059 * @sg: The array of scatter/gather entries 2060 * @nents: The number of scatter/gather entries 2061 * @direction: The direction of the DMA 2062 */ 2063 static inline void ib_dma_unmap_sg(struct ib_device *dev, 2064 struct scatterlist *sg, int nents, 2065 enum dma_data_direction direction) 2066 { 2067 if (dev->dma_ops) 2068 dev->dma_ops->unmap_sg(dev, sg, nents, direction); 2069 else 2070 dma_unmap_sg(dev->dma_device, sg, nents, direction); 2071 } 2072 2073 static inline int ib_dma_map_sg_attrs(struct ib_device *dev, 2074 struct scatterlist *sg, int nents, 2075 enum dma_data_direction direction, 2076 struct dma_attrs *attrs) 2077 { 2078 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, attrs); 2079 } 2080 2081 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev, 2082 struct scatterlist *sg, int nents, 2083 enum dma_data_direction direction, 2084 struct dma_attrs *attrs) 2085 { 2086 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, attrs); 2087 } 2088 /** 2089 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry 2090 * @dev: The device for which the DMA addresses were created 2091 * @sg: The scatter/gather entry 2092 */ 2093 static inline u64 ib_sg_dma_address(struct ib_device *dev, 2094 struct scatterlist *sg) 2095 { 2096 if (dev->dma_ops) 2097 return dev->dma_ops->dma_address(dev, sg); 2098 return sg_dma_address(sg); 2099 } 2100 2101 /** 2102 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry 2103 * @dev: The device for which the DMA addresses were created 2104 * @sg: The scatter/gather entry 2105 */ 2106 static inline unsigned int ib_sg_dma_len(struct ib_device *dev, 2107 struct scatterlist *sg) 2108 { 2109 if (dev->dma_ops) 2110 return dev->dma_ops->dma_len(dev, sg); 2111 return sg_dma_len(sg); 2112 } 2113 2114 /** 2115 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU 2116 * @dev: The device for which the DMA address was created 2117 * @addr: The DMA address 2118 * @size: The size of the region in bytes 2119 * @dir: The direction of the DMA 2120 */ 2121 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev, 2122 u64 addr, 2123 size_t size, 2124 enum dma_data_direction dir) 2125 { 2126 if (dev->dma_ops) 2127 dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir); 2128 else 2129 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir); 2130 } 2131 2132 /** 2133 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device 2134 * @dev: The device for which the DMA address was created 2135 * @addr: The DMA address 2136 * @size: The size of the region in bytes 2137 * @dir: The direction of the DMA 2138 */ 2139 static inline void ib_dma_sync_single_for_device(struct ib_device *dev, 2140 u64 addr, 2141 size_t size, 2142 enum dma_data_direction dir) 2143 { 2144 if (dev->dma_ops) 2145 dev->dma_ops->sync_single_for_device(dev, addr, size, dir); 2146 else 2147 dma_sync_single_for_device(dev->dma_device, addr, size, dir); 2148 } 2149 2150 /** 2151 * ib_dma_alloc_coherent - Allocate memory and map it for DMA 2152 * @dev: The device for which the DMA address is requested 2153 * @size: The size of the region to allocate in bytes 2154 * @dma_handle: A pointer for returning the DMA address of the region 2155 * @flag: memory allocator flags 2156 */ 2157 static inline void *ib_dma_alloc_coherent(struct ib_device *dev, 2158 size_t size, 2159 u64 *dma_handle, 2160 gfp_t flag) 2161 { 2162 if (dev->dma_ops) 2163 return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag); 2164 else { 2165 dma_addr_t handle; 2166 void *ret; 2167 2168 ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag); 2169 *dma_handle = handle; 2170 return ret; 2171 } 2172 } 2173 2174 /** 2175 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent() 2176 * @dev: The device for which the DMA addresses were allocated 2177 * @size: The size of the region 2178 * @cpu_addr: the address returned by ib_dma_alloc_coherent() 2179 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent() 2180 */ 2181 static inline void ib_dma_free_coherent(struct ib_device *dev, 2182 size_t size, void *cpu_addr, 2183 u64 dma_handle) 2184 { 2185 if (dev->dma_ops) 2186 dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle); 2187 else 2188 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle); 2189 } 2190 2191 /** 2192 * ib_reg_phys_mr - Prepares a virtually addressed memory region for use 2193 * by an HCA. 2194 * @pd: The protection domain associated assigned to the registered region. 2195 * @phys_buf_array: Specifies a list of physical buffers to use in the 2196 * memory region. 2197 * @num_phys_buf: Specifies the size of the phys_buf_array. 2198 * @mr_access_flags: Specifies the memory access rights. 2199 * @iova_start: The offset of the region's starting I/O virtual address. 2200 */ 2201 struct ib_mr *ib_reg_phys_mr(struct ib_pd *pd, 2202 struct ib_phys_buf *phys_buf_array, 2203 int num_phys_buf, 2204 int mr_access_flags, 2205 u64 *iova_start); 2206 2207 /** 2208 * ib_rereg_phys_mr - Modifies the attributes of an existing memory region. 2209 * Conceptually, this call performs the functions deregister memory region 2210 * followed by register physical memory region. Where possible, 2211 * resources are reused instead of deallocated and reallocated. 2212 * @mr: The memory region to modify. 2213 * @mr_rereg_mask: A bit-mask used to indicate which of the following 2214 * properties of the memory region are being modified. 2215 * @pd: If %IB_MR_REREG_PD is set in mr_rereg_mask, this field specifies 2216 * the new protection domain to associated with the memory region, 2217 * otherwise, this parameter is ignored. 2218 * @phys_buf_array: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this 2219 * field specifies a list of physical buffers to use in the new 2220 * translation, otherwise, this parameter is ignored. 2221 * @num_phys_buf: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this 2222 * field specifies the size of the phys_buf_array, otherwise, this 2223 * parameter is ignored. 2224 * @mr_access_flags: If %IB_MR_REREG_ACCESS is set in mr_rereg_mask, this 2225 * field specifies the new memory access rights, otherwise, this 2226 * parameter is ignored. 2227 * @iova_start: The offset of the region's starting I/O virtual address. 2228 */ 2229 int ib_rereg_phys_mr(struct ib_mr *mr, 2230 int mr_rereg_mask, 2231 struct ib_pd *pd, 2232 struct ib_phys_buf *phys_buf_array, 2233 int num_phys_buf, 2234 int mr_access_flags, 2235 u64 *iova_start); 2236 2237 /** 2238 * ib_query_mr - Retrieves information about a specific memory region. 2239 * @mr: The memory region to retrieve information about. 2240 * @mr_attr: The attributes of the specified memory region. 2241 */ 2242 int ib_query_mr(struct ib_mr *mr, struct ib_mr_attr *mr_attr); 2243 2244 /** 2245 * ib_dereg_mr - Deregisters a memory region and removes it from the 2246 * HCA translation table. 2247 * @mr: The memory region to deregister. 2248 * 2249 * This function can fail, if the memory region has memory windows bound to it. 2250 */ 2251 int ib_dereg_mr(struct ib_mr *mr); 2252 2253 /** 2254 * ib_alloc_fast_reg_mr - Allocates memory region usable with the 2255 * IB_WR_FAST_REG_MR send work request. 2256 * @pd: The protection domain associated with the region. 2257 * @max_page_list_len: requested max physical buffer list length to be 2258 * used with fast register work requests for this MR. 2259 */ 2260 struct ib_mr *ib_alloc_fast_reg_mr(struct ib_pd *pd, int max_page_list_len); 2261 2262 /** 2263 * ib_alloc_fast_reg_page_list - Allocates a page list array 2264 * @device - ib device pointer. 2265 * @page_list_len - size of the page list array to be allocated. 2266 * 2267 * This allocates and returns a struct ib_fast_reg_page_list * and a 2268 * page_list array that is at least page_list_len in size. The actual 2269 * size is returned in max_page_list_len. The caller is responsible 2270 * for initializing the contents of the page_list array before posting 2271 * a send work request with the IB_WC_FAST_REG_MR opcode. 2272 * 2273 * The page_list array entries must be translated using one of the 2274 * ib_dma_*() functions just like the addresses passed to 2275 * ib_map_phys_fmr(). Once the ib_post_send() is issued, the struct 2276 * ib_fast_reg_page_list must not be modified by the caller until the 2277 * IB_WC_FAST_REG_MR work request completes. 2278 */ 2279 struct ib_fast_reg_page_list *ib_alloc_fast_reg_page_list( 2280 struct ib_device *device, int page_list_len); 2281 2282 /** 2283 * ib_free_fast_reg_page_list - Deallocates a previously allocated 2284 * page list array. 2285 * @page_list - struct ib_fast_reg_page_list pointer to be deallocated. 2286 */ 2287 void ib_free_fast_reg_page_list(struct ib_fast_reg_page_list *page_list); 2288 2289 /** 2290 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR 2291 * R_Key and L_Key. 2292 * @mr - struct ib_mr pointer to be updated. 2293 * @newkey - new key to be used. 2294 */ 2295 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey) 2296 { 2297 mr->lkey = (mr->lkey & 0xffffff00) | newkey; 2298 mr->rkey = (mr->rkey & 0xffffff00) | newkey; 2299 } 2300 2301 /** 2302 * ib_inc_rkey - increments the key portion of the given rkey. Can be used 2303 * for calculating a new rkey for type 2 memory windows. 2304 * @rkey - the rkey to increment. 2305 */ 2306 static inline u32 ib_inc_rkey(u32 rkey) 2307 { 2308 const u32 mask = 0x000000ff; 2309 return ((rkey + 1) & mask) | (rkey & ~mask); 2310 } 2311 2312 /** 2313 * ib_alloc_mw - Allocates a memory window. 2314 * @pd: The protection domain associated with the memory window. 2315 * @type: The type of the memory window (1 or 2). 2316 */ 2317 struct ib_mw *ib_alloc_mw(struct ib_pd *pd, enum ib_mw_type type); 2318 2319 /** 2320 * ib_bind_mw - Posts a work request to the send queue of the specified 2321 * QP, which binds the memory window to the given address range and 2322 * remote access attributes. 2323 * @qp: QP to post the bind work request on. 2324 * @mw: The memory window to bind. 2325 * @mw_bind: Specifies information about the memory window, including 2326 * its address range, remote access rights, and associated memory region. 2327 * 2328 * If there is no immediate error, the function will update the rkey member 2329 * of the mw parameter to its new value. The bind operation can still fail 2330 * asynchronously. 2331 */ 2332 static inline int ib_bind_mw(struct ib_qp *qp, 2333 struct ib_mw *mw, 2334 struct ib_mw_bind *mw_bind) 2335 { 2336 /* XXX reference counting in corresponding MR? */ 2337 return mw->device->bind_mw ? 2338 mw->device->bind_mw(qp, mw, mw_bind) : 2339 -ENOSYS; 2340 } 2341 2342 /** 2343 * ib_dealloc_mw - Deallocates a memory window. 2344 * @mw: The memory window to deallocate. 2345 */ 2346 int ib_dealloc_mw(struct ib_mw *mw); 2347 2348 /** 2349 * ib_alloc_fmr - Allocates a unmapped fast memory region. 2350 * @pd: The protection domain associated with the unmapped region. 2351 * @mr_access_flags: Specifies the memory access rights. 2352 * @fmr_attr: Attributes of the unmapped region. 2353 * 2354 * A fast memory region must be mapped before it can be used as part of 2355 * a work request. 2356 */ 2357 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd, 2358 int mr_access_flags, 2359 struct ib_fmr_attr *fmr_attr); 2360 2361 /** 2362 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region. 2363 * @fmr: The fast memory region to associate with the pages. 2364 * @page_list: An array of physical pages to map to the fast memory region. 2365 * @list_len: The number of pages in page_list. 2366 * @iova: The I/O virtual address to use with the mapped region. 2367 */ 2368 static inline int ib_map_phys_fmr(struct ib_fmr *fmr, 2369 u64 *page_list, int list_len, 2370 u64 iova) 2371 { 2372 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova); 2373 } 2374 2375 /** 2376 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions. 2377 * @fmr_list: A linked list of fast memory regions to unmap. 2378 */ 2379 int ib_unmap_fmr(struct list_head *fmr_list); 2380 2381 /** 2382 * ib_dealloc_fmr - Deallocates a fast memory region. 2383 * @fmr: The fast memory region to deallocate. 2384 */ 2385 int ib_dealloc_fmr(struct ib_fmr *fmr); 2386 2387 /** 2388 * ib_attach_mcast - Attaches the specified QP to a multicast group. 2389 * @qp: QP to attach to the multicast group. The QP must be type 2390 * IB_QPT_UD. 2391 * @gid: Multicast group GID. 2392 * @lid: Multicast group LID in host byte order. 2393 * 2394 * In order to send and receive multicast packets, subnet 2395 * administration must have created the multicast group and configured 2396 * the fabric appropriately. The port associated with the specified 2397 * QP must also be a member of the multicast group. 2398 */ 2399 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid); 2400 2401 /** 2402 * ib_detach_mcast - Detaches the specified QP from a multicast group. 2403 * @qp: QP to detach from the multicast group. 2404 * @gid: Multicast group GID. 2405 * @lid: Multicast group LID in host byte order. 2406 */ 2407 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid); 2408 2409 /** 2410 * ib_alloc_xrcd - Allocates an XRC domain. 2411 * @device: The device on which to allocate the XRC domain. 2412 */ 2413 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device); 2414 2415 /** 2416 * ib_dealloc_xrcd - Deallocates an XRC domain. 2417 * @xrcd: The XRC domain to deallocate. 2418 */ 2419 int ib_dealloc_xrcd(struct ib_xrcd *xrcd); 2420 2421 struct ib_flow *ib_create_flow(struct ib_qp *qp, 2422 struct ib_flow_attr *flow_attr, int domain); 2423 int ib_destroy_flow(struct ib_flow *flow_id); 2424 2425 static inline int ib_check_mr_access(int flags) 2426 { 2427 /* 2428 * Local write permission is required if remote write or 2429 * remote atomic permission is also requested. 2430 */ 2431 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) && 2432 !(flags & IB_ACCESS_LOCAL_WRITE)) 2433 return -EINVAL; 2434 2435 return 0; 2436 } 2437 2438 #endif /* IB_VERBS_H */ 2439