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