ib_verbs.h revision 331784
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 * $FreeBSD: stable/11/sys/ofed/include/rdma/ib_verbs.h 331784 2018-03-30 18:36:44Z hselasky $ 39 */ 40 41#if !defined(IB_VERBS_H) 42#define IB_VERBS_H 43 44#include <linux/types.h> 45#include <linux/device.h> 46#include <linux/mm.h> 47#include <linux/dma-mapping.h> 48#include <linux/kref.h> 49#include <linux/list.h> 50#include <linux/rwsem.h> 51#include <linux/scatterlist.h> 52#include <linux/workqueue.h> 53#include <linux/socket.h> 54#include <linux/if_ether.h> 55#include <net/ipv6.h> 56#include <net/ip.h> 57#include <linux/string.h> 58#include <linux/slab.h> 59#include <linux/rcupdate.h> 60#include <linux/netdevice.h> 61#include <netinet/ip.h> 62 63#include <asm/atomic.h> 64#include <asm/uaccess.h> 65 66struct ifla_vf_info; 67struct ifla_vf_stats; 68 69extern struct workqueue_struct *ib_wq; 70extern struct workqueue_struct *ib_comp_wq; 71 72union ib_gid { 73 u8 raw[16]; 74 struct { 75 __be64 subnet_prefix; 76 __be64 interface_id; 77 } global; 78}; 79 80extern union ib_gid zgid; 81 82enum ib_gid_type { 83 /* If link layer is Ethernet, this is RoCE V1 */ 84 IB_GID_TYPE_IB = 0, 85 IB_GID_TYPE_ROCE = 0, 86 IB_GID_TYPE_ROCE_UDP_ENCAP = 1, 87 IB_GID_TYPE_SIZE 88}; 89 90#define ROCE_V2_UDP_DPORT 4791 91struct ib_gid_attr { 92 enum ib_gid_type gid_type; 93 struct net_device *ndev; 94}; 95 96enum rdma_node_type { 97 /* IB values map to NodeInfo:NodeType. */ 98 RDMA_NODE_IB_CA = 1, 99 RDMA_NODE_IB_SWITCH, 100 RDMA_NODE_IB_ROUTER, 101 RDMA_NODE_RNIC, 102 RDMA_NODE_USNIC, 103 RDMA_NODE_USNIC_UDP, 104}; 105 106enum { 107 /* set the local administered indication */ 108 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2, 109}; 110 111enum rdma_transport_type { 112 RDMA_TRANSPORT_IB, 113 RDMA_TRANSPORT_IWARP, 114 RDMA_TRANSPORT_USNIC, 115 RDMA_TRANSPORT_USNIC_UDP 116}; 117 118enum rdma_protocol_type { 119 RDMA_PROTOCOL_IB, 120 RDMA_PROTOCOL_IBOE, 121 RDMA_PROTOCOL_IWARP, 122 RDMA_PROTOCOL_USNIC_UDP 123}; 124 125__attribute_const__ enum rdma_transport_type 126rdma_node_get_transport(enum rdma_node_type node_type); 127 128enum rdma_network_type { 129 RDMA_NETWORK_IB, 130 RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB, 131 RDMA_NETWORK_IPV4, 132 RDMA_NETWORK_IPV6 133}; 134 135static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type) 136{ 137 if (network_type == RDMA_NETWORK_IPV4 || 138 network_type == RDMA_NETWORK_IPV6) 139 return IB_GID_TYPE_ROCE_UDP_ENCAP; 140 141 /* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */ 142 return IB_GID_TYPE_IB; 143} 144 145static inline enum rdma_network_type ib_gid_to_network_type(enum ib_gid_type gid_type, 146 union ib_gid *gid) 147{ 148 if (gid_type == IB_GID_TYPE_IB) 149 return RDMA_NETWORK_IB; 150 151 if (ipv6_addr_v4mapped((struct in6_addr *)gid)) 152 return RDMA_NETWORK_IPV4; 153 else 154 return RDMA_NETWORK_IPV6; 155} 156 157enum rdma_link_layer { 158 IB_LINK_LAYER_UNSPECIFIED, 159 IB_LINK_LAYER_INFINIBAND, 160 IB_LINK_LAYER_ETHERNET, 161}; 162 163enum ib_device_cap_flags { 164 IB_DEVICE_RESIZE_MAX_WR = (1 << 0), 165 IB_DEVICE_BAD_PKEY_CNTR = (1 << 1), 166 IB_DEVICE_BAD_QKEY_CNTR = (1 << 2), 167 IB_DEVICE_RAW_MULTI = (1 << 3), 168 IB_DEVICE_AUTO_PATH_MIG = (1 << 4), 169 IB_DEVICE_CHANGE_PHY_PORT = (1 << 5), 170 IB_DEVICE_UD_AV_PORT_ENFORCE = (1 << 6), 171 IB_DEVICE_CURR_QP_STATE_MOD = (1 << 7), 172 IB_DEVICE_SHUTDOWN_PORT = (1 << 8), 173 IB_DEVICE_INIT_TYPE = (1 << 9), 174 IB_DEVICE_PORT_ACTIVE_EVENT = (1 << 10), 175 IB_DEVICE_SYS_IMAGE_GUID = (1 << 11), 176 IB_DEVICE_RC_RNR_NAK_GEN = (1 << 12), 177 IB_DEVICE_SRQ_RESIZE = (1 << 13), 178 IB_DEVICE_N_NOTIFY_CQ = (1 << 14), 179 180 /* 181 * This device supports a per-device lkey or stag that can be 182 * used without performing a memory registration for the local 183 * memory. Note that ULPs should never check this flag, but 184 * instead of use the local_dma_lkey flag in the ib_pd structure, 185 * which will always contain a usable lkey. 186 */ 187 IB_DEVICE_LOCAL_DMA_LKEY = (1 << 15), 188 IB_DEVICE_RESERVED /* old SEND_W_INV */ = (1 << 16), 189 IB_DEVICE_MEM_WINDOW = (1 << 17), 190 /* 191 * Devices should set IB_DEVICE_UD_IP_SUM if they support 192 * insertion of UDP and TCP checksum on outgoing UD IPoIB 193 * messages and can verify the validity of checksum for 194 * incoming messages. Setting this flag implies that the 195 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode. 196 */ 197 IB_DEVICE_UD_IP_CSUM = (1 << 18), 198 IB_DEVICE_UD_TSO = (1 << 19), 199 IB_DEVICE_XRC = (1 << 20), 200 201 /* 202 * This device supports the IB "base memory management extension", 203 * which includes support for fast registrations (IB_WR_REG_MR, 204 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should 205 * also be set by any iWarp device which must support FRs to comply 206 * to the iWarp verbs spec. iWarp devices also support the 207 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the 208 * stag. 209 */ 210 IB_DEVICE_MEM_MGT_EXTENSIONS = (1 << 21), 211 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1 << 22), 212 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1 << 23), 213 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1 << 24), 214 IB_DEVICE_RC_IP_CSUM = (1 << 25), 215 IB_DEVICE_RAW_IP_CSUM = (1 << 26), 216 /* 217 * Devices should set IB_DEVICE_CROSS_CHANNEL if they 218 * support execution of WQEs that involve synchronization 219 * of I/O operations with single completion queue managed 220 * by hardware. 221 */ 222 IB_DEVICE_CROSS_CHANNEL = (1 << 27), 223 IB_DEVICE_MANAGED_FLOW_STEERING = (1 << 29), 224 IB_DEVICE_SIGNATURE_HANDOVER = (1 << 30), 225 IB_DEVICE_ON_DEMAND_PAGING = (1ULL << 31), 226 IB_DEVICE_SG_GAPS_REG = (1ULL << 32), 227 IB_DEVICE_VIRTUAL_FUNCTION = (1ULL << 33), 228 IB_DEVICE_RAW_SCATTER_FCS = (1ULL << 34), 229}; 230 231enum ib_signature_prot_cap { 232 IB_PROT_T10DIF_TYPE_1 = 1, 233 IB_PROT_T10DIF_TYPE_2 = 1 << 1, 234 IB_PROT_T10DIF_TYPE_3 = 1 << 2, 235}; 236 237enum ib_signature_guard_cap { 238 IB_GUARD_T10DIF_CRC = 1, 239 IB_GUARD_T10DIF_CSUM = 1 << 1, 240}; 241 242enum ib_atomic_cap { 243 IB_ATOMIC_NONE, 244 IB_ATOMIC_HCA, 245 IB_ATOMIC_GLOB 246}; 247 248enum ib_odp_general_cap_bits { 249 IB_ODP_SUPPORT = 1 << 0, 250}; 251 252enum ib_odp_transport_cap_bits { 253 IB_ODP_SUPPORT_SEND = 1 << 0, 254 IB_ODP_SUPPORT_RECV = 1 << 1, 255 IB_ODP_SUPPORT_WRITE = 1 << 2, 256 IB_ODP_SUPPORT_READ = 1 << 3, 257 IB_ODP_SUPPORT_ATOMIC = 1 << 4, 258}; 259 260struct ib_odp_caps { 261 uint64_t general_caps; 262 struct { 263 uint32_t rc_odp_caps; 264 uint32_t uc_odp_caps; 265 uint32_t ud_odp_caps; 266 } per_transport_caps; 267}; 268 269struct ib_rss_caps { 270 /* Corresponding bit will be set if qp type from 271 * 'enum ib_qp_type' is supported, e.g. 272 * supported_qpts |= 1 << IB_QPT_UD 273 */ 274 u32 supported_qpts; 275 u32 max_rwq_indirection_tables; 276 u32 max_rwq_indirection_table_size; 277}; 278 279enum ib_cq_creation_flags { 280 IB_CQ_FLAGS_TIMESTAMP_COMPLETION = 1 << 0, 281 IB_CQ_FLAGS_IGNORE_OVERRUN = 1 << 1, 282}; 283 284struct ib_cq_init_attr { 285 unsigned int cqe; 286 int comp_vector; 287 u32 flags; 288}; 289 290struct ib_device_attr { 291 u64 fw_ver; 292 __be64 sys_image_guid; 293 u64 max_mr_size; 294 u64 page_size_cap; 295 u32 vendor_id; 296 u32 vendor_part_id; 297 u32 hw_ver; 298 int max_qp; 299 int max_qp_wr; 300 u64 device_cap_flags; 301 int max_sge; 302 int max_sge_rd; 303 int max_cq; 304 int max_cqe; 305 int max_mr; 306 int max_pd; 307 int max_qp_rd_atom; 308 int max_ee_rd_atom; 309 int max_res_rd_atom; 310 int max_qp_init_rd_atom; 311 int max_ee_init_rd_atom; 312 enum ib_atomic_cap atomic_cap; 313 enum ib_atomic_cap masked_atomic_cap; 314 int max_ee; 315 int max_rdd; 316 int max_mw; 317 int max_raw_ipv6_qp; 318 int max_raw_ethy_qp; 319 int max_mcast_grp; 320 int max_mcast_qp_attach; 321 int max_total_mcast_qp_attach; 322 int max_ah; 323 int max_fmr; 324 int max_map_per_fmr; 325 int max_srq; 326 int max_srq_wr; 327 int max_srq_sge; 328 unsigned int max_fast_reg_page_list_len; 329 u16 max_pkeys; 330 u8 local_ca_ack_delay; 331 int sig_prot_cap; 332 int sig_guard_cap; 333 struct ib_odp_caps odp_caps; 334 uint64_t timestamp_mask; 335 uint64_t hca_core_clock; /* in KHZ */ 336 struct ib_rss_caps rss_caps; 337 u32 max_wq_type_rq; 338}; 339 340enum ib_mtu { 341 IB_MTU_256 = 1, 342 IB_MTU_512 = 2, 343 IB_MTU_1024 = 3, 344 IB_MTU_2048 = 4, 345 IB_MTU_4096 = 5 346}; 347 348static inline int ib_mtu_enum_to_int(enum ib_mtu mtu) 349{ 350 switch (mtu) { 351 case IB_MTU_256: return 256; 352 case IB_MTU_512: return 512; 353 case IB_MTU_1024: return 1024; 354 case IB_MTU_2048: return 2048; 355 case IB_MTU_4096: return 4096; 356 default: return -1; 357 } 358} 359 360enum ib_port_state { 361 IB_PORT_NOP = 0, 362 IB_PORT_DOWN = 1, 363 IB_PORT_INIT = 2, 364 IB_PORT_ARMED = 3, 365 IB_PORT_ACTIVE = 4, 366 IB_PORT_ACTIVE_DEFER = 5, 367 IB_PORT_DUMMY = -1, /* force enum signed */ 368}; 369 370enum ib_port_cap_flags { 371 IB_PORT_SM = 1 << 1, 372 IB_PORT_NOTICE_SUP = 1 << 2, 373 IB_PORT_TRAP_SUP = 1 << 3, 374 IB_PORT_OPT_IPD_SUP = 1 << 4, 375 IB_PORT_AUTO_MIGR_SUP = 1 << 5, 376 IB_PORT_SL_MAP_SUP = 1 << 6, 377 IB_PORT_MKEY_NVRAM = 1 << 7, 378 IB_PORT_PKEY_NVRAM = 1 << 8, 379 IB_PORT_LED_INFO_SUP = 1 << 9, 380 IB_PORT_SM_DISABLED = 1 << 10, 381 IB_PORT_SYS_IMAGE_GUID_SUP = 1 << 11, 382 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP = 1 << 12, 383 IB_PORT_EXTENDED_SPEEDS_SUP = 1 << 14, 384 IB_PORT_CM_SUP = 1 << 16, 385 IB_PORT_SNMP_TUNNEL_SUP = 1 << 17, 386 IB_PORT_REINIT_SUP = 1 << 18, 387 IB_PORT_DEVICE_MGMT_SUP = 1 << 19, 388 IB_PORT_VENDOR_CLASS_SUP = 1 << 20, 389 IB_PORT_DR_NOTICE_SUP = 1 << 21, 390 IB_PORT_CAP_MASK_NOTICE_SUP = 1 << 22, 391 IB_PORT_BOOT_MGMT_SUP = 1 << 23, 392 IB_PORT_LINK_LATENCY_SUP = 1 << 24, 393 IB_PORT_CLIENT_REG_SUP = 1 << 25, 394 IB_PORT_IP_BASED_GIDS = 1 << 26, 395}; 396 397enum ib_port_width { 398 IB_WIDTH_1X = 1, 399 IB_WIDTH_4X = 2, 400 IB_WIDTH_8X = 4, 401 IB_WIDTH_12X = 8 402}; 403 404static inline int ib_width_enum_to_int(enum ib_port_width width) 405{ 406 switch (width) { 407 case IB_WIDTH_1X: return 1; 408 case IB_WIDTH_4X: return 4; 409 case IB_WIDTH_8X: return 8; 410 case IB_WIDTH_12X: return 12; 411 default: return -1; 412 } 413} 414 415enum ib_port_speed { 416 IB_SPEED_SDR = 1, 417 IB_SPEED_DDR = 2, 418 IB_SPEED_QDR = 4, 419 IB_SPEED_FDR10 = 8, 420 IB_SPEED_FDR = 16, 421 IB_SPEED_EDR = 32 422}; 423 424/** 425 * struct rdma_hw_stats 426 * @timestamp - Used by the core code to track when the last update was 427 * @lifespan - Used by the core code to determine how old the counters 428 * should be before being updated again. Stored in jiffies, defaults 429 * to 10 milliseconds, drivers can override the default be specifying 430 * their own value during their allocation routine. 431 * @name - Array of pointers to static names used for the counters in 432 * directory. 433 * @num_counters - How many hardware counters there are. If name is 434 * shorter than this number, a kernel oops will result. Driver authors 435 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters) 436 * in their code to prevent this. 437 * @value - Array of u64 counters that are accessed by the sysfs code and 438 * filled in by the drivers get_stats routine 439 */ 440struct rdma_hw_stats { 441 unsigned long timestamp; 442 unsigned long lifespan; 443 const char * const *names; 444 int num_counters; 445 u64 value[]; 446}; 447 448#define RDMA_HW_STATS_DEFAULT_LIFESPAN 10 449/** 450 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct 451 * for drivers. 452 * @names - Array of static const char * 453 * @num_counters - How many elements in array 454 * @lifespan - How many milliseconds between updates 455 */ 456static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct( 457 const char * const *names, int num_counters, 458 unsigned long lifespan) 459{ 460 struct rdma_hw_stats *stats; 461 462 stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64), 463 GFP_KERNEL); 464 if (!stats) 465 return NULL; 466 stats->names = names; 467 stats->num_counters = num_counters; 468 stats->lifespan = msecs_to_jiffies(lifespan); 469 470 return stats; 471} 472 473 474/* Define bits for the various functionality this port needs to be supported by 475 * the core. 476 */ 477/* Management 0x00000FFF */ 478#define RDMA_CORE_CAP_IB_MAD 0x00000001 479#define RDMA_CORE_CAP_IB_SMI 0x00000002 480#define RDMA_CORE_CAP_IB_CM 0x00000004 481#define RDMA_CORE_CAP_IW_CM 0x00000008 482#define RDMA_CORE_CAP_IB_SA 0x00000010 483#define RDMA_CORE_CAP_OPA_MAD 0x00000020 484 485/* Address format 0x000FF000 */ 486#define RDMA_CORE_CAP_AF_IB 0x00001000 487#define RDMA_CORE_CAP_ETH_AH 0x00002000 488 489/* Protocol 0xFFF00000 */ 490#define RDMA_CORE_CAP_PROT_IB 0x00100000 491#define RDMA_CORE_CAP_PROT_ROCE 0x00200000 492#define RDMA_CORE_CAP_PROT_IWARP 0x00400000 493#define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000 494 495#define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \ 496 | RDMA_CORE_CAP_IB_MAD \ 497 | RDMA_CORE_CAP_IB_SMI \ 498 | RDMA_CORE_CAP_IB_CM \ 499 | RDMA_CORE_CAP_IB_SA \ 500 | RDMA_CORE_CAP_AF_IB) 501#define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \ 502 | RDMA_CORE_CAP_IB_MAD \ 503 | RDMA_CORE_CAP_IB_CM \ 504 | RDMA_CORE_CAP_AF_IB \ 505 | RDMA_CORE_CAP_ETH_AH) 506#define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \ 507 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \ 508 | RDMA_CORE_CAP_IB_MAD \ 509 | RDMA_CORE_CAP_IB_CM \ 510 | RDMA_CORE_CAP_AF_IB \ 511 | RDMA_CORE_CAP_ETH_AH) 512#define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \ 513 | RDMA_CORE_CAP_IW_CM) 514#define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \ 515 | RDMA_CORE_CAP_OPA_MAD) 516 517struct ib_port_attr { 518 u64 subnet_prefix; 519 enum ib_port_state state; 520 enum ib_mtu max_mtu; 521 enum ib_mtu active_mtu; 522 int gid_tbl_len; 523 u32 port_cap_flags; 524 u32 max_msg_sz; 525 u32 bad_pkey_cntr; 526 u32 qkey_viol_cntr; 527 u16 pkey_tbl_len; 528 u16 lid; 529 u16 sm_lid; 530 u8 lmc; 531 u8 max_vl_num; 532 u8 sm_sl; 533 u8 subnet_timeout; 534 u8 init_type_reply; 535 u8 active_width; 536 u8 active_speed; 537 u8 phys_state; 538 bool grh_required; 539}; 540 541enum ib_device_modify_flags { 542 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0, 543 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1 544}; 545 546#define IB_DEVICE_NODE_DESC_MAX 64 547 548struct ib_device_modify { 549 u64 sys_image_guid; 550 char node_desc[IB_DEVICE_NODE_DESC_MAX]; 551}; 552 553enum ib_port_modify_flags { 554 IB_PORT_SHUTDOWN = 1, 555 IB_PORT_INIT_TYPE = (1<<2), 556 IB_PORT_RESET_QKEY_CNTR = (1<<3) 557}; 558 559struct ib_port_modify { 560 u32 set_port_cap_mask; 561 u32 clr_port_cap_mask; 562 u8 init_type; 563}; 564 565enum ib_event_type { 566 IB_EVENT_CQ_ERR, 567 IB_EVENT_QP_FATAL, 568 IB_EVENT_QP_REQ_ERR, 569 IB_EVENT_QP_ACCESS_ERR, 570 IB_EVENT_COMM_EST, 571 IB_EVENT_SQ_DRAINED, 572 IB_EVENT_PATH_MIG, 573 IB_EVENT_PATH_MIG_ERR, 574 IB_EVENT_DEVICE_FATAL, 575 IB_EVENT_PORT_ACTIVE, 576 IB_EVENT_PORT_ERR, 577 IB_EVENT_LID_CHANGE, 578 IB_EVENT_PKEY_CHANGE, 579 IB_EVENT_SM_CHANGE, 580 IB_EVENT_SRQ_ERR, 581 IB_EVENT_SRQ_LIMIT_REACHED, 582 IB_EVENT_QP_LAST_WQE_REACHED, 583 IB_EVENT_CLIENT_REREGISTER, 584 IB_EVENT_GID_CHANGE, 585 IB_EVENT_WQ_FATAL, 586}; 587 588const char *__attribute_const__ ib_event_msg(enum ib_event_type event); 589 590struct ib_event { 591 struct ib_device *device; 592 union { 593 struct ib_cq *cq; 594 struct ib_qp *qp; 595 struct ib_srq *srq; 596 struct ib_wq *wq; 597 u8 port_num; 598 } element; 599 enum ib_event_type event; 600}; 601 602struct ib_event_handler { 603 struct ib_device *device; 604 void (*handler)(struct ib_event_handler *, struct ib_event *); 605 struct list_head list; 606}; 607 608#define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \ 609 do { \ 610 (_ptr)->device = _device; \ 611 (_ptr)->handler = _handler; \ 612 INIT_LIST_HEAD(&(_ptr)->list); \ 613 } while (0) 614 615struct ib_global_route { 616 union ib_gid dgid; 617 u32 flow_label; 618 u8 sgid_index; 619 u8 hop_limit; 620 u8 traffic_class; 621}; 622 623struct ib_grh { 624 __be32 version_tclass_flow; 625 __be16 paylen; 626 u8 next_hdr; 627 u8 hop_limit; 628 union ib_gid sgid; 629 union ib_gid dgid; 630}; 631 632union rdma_network_hdr { 633 struct ib_grh ibgrh; 634 struct { 635 /* The IB spec states that if it's IPv4, the header 636 * is located in the last 20 bytes of the header. 637 */ 638 u8 reserved[20]; 639 struct ip roce4grh; 640 }; 641}; 642 643enum { 644 IB_MULTICAST_QPN = 0xffffff 645}; 646 647#define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF) 648#define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000) 649 650enum ib_ah_flags { 651 IB_AH_GRH = 1 652}; 653 654enum ib_rate { 655 IB_RATE_PORT_CURRENT = 0, 656 IB_RATE_2_5_GBPS = 2, 657 IB_RATE_5_GBPS = 5, 658 IB_RATE_10_GBPS = 3, 659 IB_RATE_20_GBPS = 6, 660 IB_RATE_30_GBPS = 4, 661 IB_RATE_40_GBPS = 7, 662 IB_RATE_60_GBPS = 8, 663 IB_RATE_80_GBPS = 9, 664 IB_RATE_120_GBPS = 10, 665 IB_RATE_14_GBPS = 11, 666 IB_RATE_56_GBPS = 12, 667 IB_RATE_112_GBPS = 13, 668 IB_RATE_168_GBPS = 14, 669 IB_RATE_25_GBPS = 15, 670 IB_RATE_100_GBPS = 16, 671 IB_RATE_200_GBPS = 17, 672 IB_RATE_300_GBPS = 18 673}; 674 675/** 676 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the 677 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be 678 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec. 679 * @rate: rate to convert. 680 */ 681__attribute_const__ int ib_rate_to_mult(enum ib_rate rate); 682 683/** 684 * ib_rate_to_mbps - Convert the IB rate enum to Mbps. 685 * For example, IB_RATE_2_5_GBPS will be converted to 2500. 686 * @rate: rate to convert. 687 */ 688__attribute_const__ int ib_rate_to_mbps(enum ib_rate rate); 689 690 691/** 692 * enum ib_mr_type - memory region type 693 * @IB_MR_TYPE_MEM_REG: memory region that is used for 694 * normal registration 695 * @IB_MR_TYPE_SIGNATURE: memory region that is used for 696 * signature operations (data-integrity 697 * capable regions) 698 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to 699 * register any arbitrary sg lists (without 700 * the normal mr constraints - see 701 * ib_map_mr_sg) 702 */ 703enum ib_mr_type { 704 IB_MR_TYPE_MEM_REG, 705 IB_MR_TYPE_SIGNATURE, 706 IB_MR_TYPE_SG_GAPS, 707}; 708 709/** 710 * Signature types 711 * IB_SIG_TYPE_NONE: Unprotected. 712 * IB_SIG_TYPE_T10_DIF: Type T10-DIF 713 */ 714enum ib_signature_type { 715 IB_SIG_TYPE_NONE, 716 IB_SIG_TYPE_T10_DIF, 717}; 718 719/** 720 * Signature T10-DIF block-guard types 721 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules. 722 * IB_T10DIF_CSUM: Corresponds to IP checksum rules. 723 */ 724enum ib_t10_dif_bg_type { 725 IB_T10DIF_CRC, 726 IB_T10DIF_CSUM 727}; 728 729/** 730 * struct ib_t10_dif_domain - Parameters specific for T10-DIF 731 * domain. 732 * @bg_type: T10-DIF block guard type (CRC|CSUM) 733 * @pi_interval: protection information interval. 734 * @bg: seed of guard computation. 735 * @app_tag: application tag of guard block 736 * @ref_tag: initial guard block reference tag. 737 * @ref_remap: Indicate wethear the reftag increments each block 738 * @app_escape: Indicate to skip block check if apptag=0xffff 739 * @ref_escape: Indicate to skip block check if reftag=0xffffffff 740 * @apptag_check_mask: check bitmask of application tag. 741 */ 742struct ib_t10_dif_domain { 743 enum ib_t10_dif_bg_type bg_type; 744 u16 pi_interval; 745 u16 bg; 746 u16 app_tag; 747 u32 ref_tag; 748 bool ref_remap; 749 bool app_escape; 750 bool ref_escape; 751 u16 apptag_check_mask; 752}; 753 754/** 755 * struct ib_sig_domain - Parameters for signature domain 756 * @sig_type: specific signauture type 757 * @sig: union of all signature domain attributes that may 758 * be used to set domain layout. 759 */ 760struct ib_sig_domain { 761 enum ib_signature_type sig_type; 762 union { 763 struct ib_t10_dif_domain dif; 764 } sig; 765}; 766 767/** 768 * struct ib_sig_attrs - Parameters for signature handover operation 769 * @check_mask: bitmask for signature byte check (8 bytes) 770 * @mem: memory domain layout desciptor. 771 * @wire: wire domain layout desciptor. 772 */ 773struct ib_sig_attrs { 774 u8 check_mask; 775 struct ib_sig_domain mem; 776 struct ib_sig_domain wire; 777}; 778 779enum ib_sig_err_type { 780 IB_SIG_BAD_GUARD, 781 IB_SIG_BAD_REFTAG, 782 IB_SIG_BAD_APPTAG, 783}; 784 785/** 786 * struct ib_sig_err - signature error descriptor 787 */ 788struct ib_sig_err { 789 enum ib_sig_err_type err_type; 790 u32 expected; 791 u32 actual; 792 u64 sig_err_offset; 793 u32 key; 794}; 795 796enum ib_mr_status_check { 797 IB_MR_CHECK_SIG_STATUS = 1, 798}; 799 800/** 801 * struct ib_mr_status - Memory region status container 802 * 803 * @fail_status: Bitmask of MR checks status. For each 804 * failed check a corresponding status bit is set. 805 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS 806 * failure. 807 */ 808struct ib_mr_status { 809 u32 fail_status; 810 struct ib_sig_err sig_err; 811}; 812 813/** 814 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate 815 * enum. 816 * @mult: multiple to convert. 817 */ 818__attribute_const__ enum ib_rate mult_to_ib_rate(int mult); 819 820struct ib_ah_attr { 821 struct ib_global_route grh; 822 u16 dlid; 823 u8 sl; 824 u8 src_path_bits; 825 u8 static_rate; 826 u8 ah_flags; 827 u8 port_num; 828 u8 dmac[ETH_ALEN]; 829}; 830 831enum ib_wc_status { 832 IB_WC_SUCCESS, 833 IB_WC_LOC_LEN_ERR, 834 IB_WC_LOC_QP_OP_ERR, 835 IB_WC_LOC_EEC_OP_ERR, 836 IB_WC_LOC_PROT_ERR, 837 IB_WC_WR_FLUSH_ERR, 838 IB_WC_MW_BIND_ERR, 839 IB_WC_BAD_RESP_ERR, 840 IB_WC_LOC_ACCESS_ERR, 841 IB_WC_REM_INV_REQ_ERR, 842 IB_WC_REM_ACCESS_ERR, 843 IB_WC_REM_OP_ERR, 844 IB_WC_RETRY_EXC_ERR, 845 IB_WC_RNR_RETRY_EXC_ERR, 846 IB_WC_LOC_RDD_VIOL_ERR, 847 IB_WC_REM_INV_RD_REQ_ERR, 848 IB_WC_REM_ABORT_ERR, 849 IB_WC_INV_EECN_ERR, 850 IB_WC_INV_EEC_STATE_ERR, 851 IB_WC_FATAL_ERR, 852 IB_WC_RESP_TIMEOUT_ERR, 853 IB_WC_GENERAL_ERR 854}; 855 856const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status); 857 858enum ib_wc_opcode { 859 IB_WC_SEND, 860 IB_WC_RDMA_WRITE, 861 IB_WC_RDMA_READ, 862 IB_WC_COMP_SWAP, 863 IB_WC_FETCH_ADD, 864 IB_WC_LSO, 865 IB_WC_LOCAL_INV, 866 IB_WC_REG_MR, 867 IB_WC_MASKED_COMP_SWAP, 868 IB_WC_MASKED_FETCH_ADD, 869/* 870 * Set value of IB_WC_RECV so consumers can test if a completion is a 871 * receive by testing (opcode & IB_WC_RECV). 872 */ 873 IB_WC_RECV = 1 << 7, 874 IB_WC_RECV_RDMA_WITH_IMM, 875 IB_WC_DUMMY = -1, /* force enum signed */ 876}; 877 878enum ib_wc_flags { 879 IB_WC_GRH = 1, 880 IB_WC_WITH_IMM = (1<<1), 881 IB_WC_WITH_INVALIDATE = (1<<2), 882 IB_WC_IP_CSUM_OK = (1<<3), 883 IB_WC_WITH_SMAC = (1<<4), 884 IB_WC_WITH_VLAN = (1<<5), 885 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6), 886}; 887 888struct ib_wc { 889 union { 890 u64 wr_id; 891 struct ib_cqe *wr_cqe; 892 }; 893 enum ib_wc_status status; 894 enum ib_wc_opcode opcode; 895 u32 vendor_err; 896 u32 byte_len; 897 struct ib_qp *qp; 898 union { 899 __be32 imm_data; 900 u32 invalidate_rkey; 901 } ex; 902 u32 src_qp; 903 int wc_flags; 904 u16 pkey_index; 905 u16 slid; 906 u8 sl; 907 u8 dlid_path_bits; 908 u8 port_num; /* valid only for DR SMPs on switches */ 909 u8 smac[ETH_ALEN]; 910 u16 vlan_id; 911 u8 network_hdr_type; 912}; 913 914enum ib_cq_notify_flags { 915 IB_CQ_SOLICITED = 1 << 0, 916 IB_CQ_NEXT_COMP = 1 << 1, 917 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP, 918 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2, 919}; 920 921enum ib_srq_type { 922 IB_SRQT_BASIC, 923 IB_SRQT_XRC 924}; 925 926enum ib_srq_attr_mask { 927 IB_SRQ_MAX_WR = 1 << 0, 928 IB_SRQ_LIMIT = 1 << 1, 929}; 930 931struct ib_srq_attr { 932 u32 max_wr; 933 u32 max_sge; 934 u32 srq_limit; 935}; 936 937struct ib_srq_init_attr { 938 void (*event_handler)(struct ib_event *, void *); 939 void *srq_context; 940 struct ib_srq_attr attr; 941 enum ib_srq_type srq_type; 942 943 union { 944 struct { 945 struct ib_xrcd *xrcd; 946 struct ib_cq *cq; 947 } xrc; 948 } ext; 949}; 950 951struct ib_qp_cap { 952 u32 max_send_wr; 953 u32 max_recv_wr; 954 u32 max_send_sge; 955 u32 max_recv_sge; 956 u32 max_inline_data; 957 958 /* 959 * Maximum number of rdma_rw_ctx structures in flight at a time. 960 * ib_create_qp() will calculate the right amount of neededed WRs 961 * and MRs based on this. 962 */ 963 u32 max_rdma_ctxs; 964}; 965 966enum ib_sig_type { 967 IB_SIGNAL_ALL_WR, 968 IB_SIGNAL_REQ_WR 969}; 970 971enum ib_qp_type { 972 /* 973 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries 974 * here (and in that order) since the MAD layer uses them as 975 * indices into a 2-entry table. 976 */ 977 IB_QPT_SMI, 978 IB_QPT_GSI, 979 980 IB_QPT_RC, 981 IB_QPT_UC, 982 IB_QPT_UD, 983 IB_QPT_RAW_IPV6, 984 IB_QPT_RAW_ETHERTYPE, 985 IB_QPT_RAW_PACKET = 8, 986 IB_QPT_XRC_INI = 9, 987 IB_QPT_XRC_TGT, 988 IB_QPT_MAX, 989 /* Reserve a range for qp types internal to the low level driver. 990 * These qp types will not be visible at the IB core layer, so the 991 * IB_QPT_MAX usages should not be affected in the core layer 992 */ 993 IB_QPT_RESERVED1 = 0x1000, 994 IB_QPT_RESERVED2, 995 IB_QPT_RESERVED3, 996 IB_QPT_RESERVED4, 997 IB_QPT_RESERVED5, 998 IB_QPT_RESERVED6, 999 IB_QPT_RESERVED7, 1000 IB_QPT_RESERVED8, 1001 IB_QPT_RESERVED9, 1002 IB_QPT_RESERVED10, 1003}; 1004 1005enum ib_qp_create_flags { 1006 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0, 1007 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1, 1008 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2, 1009 IB_QP_CREATE_MANAGED_SEND = 1 << 3, 1010 IB_QP_CREATE_MANAGED_RECV = 1 << 4, 1011 IB_QP_CREATE_NETIF_QP = 1 << 5, 1012 IB_QP_CREATE_SIGNATURE_EN = 1 << 6, 1013 IB_QP_CREATE_USE_GFP_NOIO = 1 << 7, 1014 IB_QP_CREATE_SCATTER_FCS = 1 << 8, 1015 /* reserve bits 26-31 for low level drivers' internal use */ 1016 IB_QP_CREATE_RESERVED_START = 1 << 26, 1017 IB_QP_CREATE_RESERVED_END = 1 << 31, 1018}; 1019 1020/* 1021 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler 1022 * callback to destroy the passed in QP. 1023 */ 1024 1025struct ib_qp_init_attr { 1026 void (*event_handler)(struct ib_event *, void *); 1027 void *qp_context; 1028 struct ib_cq *send_cq; 1029 struct ib_cq *recv_cq; 1030 struct ib_srq *srq; 1031 struct ib_xrcd *xrcd; /* XRC TGT QPs only */ 1032 struct ib_qp_cap cap; 1033 enum ib_sig_type sq_sig_type; 1034 enum ib_qp_type qp_type; 1035 enum ib_qp_create_flags create_flags; 1036 1037 /* 1038 * Only needed for special QP types, or when using the RW API. 1039 */ 1040 u8 port_num; 1041 struct ib_rwq_ind_table *rwq_ind_tbl; 1042}; 1043 1044struct ib_qp_open_attr { 1045 void (*event_handler)(struct ib_event *, void *); 1046 void *qp_context; 1047 u32 qp_num; 1048 enum ib_qp_type qp_type; 1049}; 1050 1051enum ib_rnr_timeout { 1052 IB_RNR_TIMER_655_36 = 0, 1053 IB_RNR_TIMER_000_01 = 1, 1054 IB_RNR_TIMER_000_02 = 2, 1055 IB_RNR_TIMER_000_03 = 3, 1056 IB_RNR_TIMER_000_04 = 4, 1057 IB_RNR_TIMER_000_06 = 5, 1058 IB_RNR_TIMER_000_08 = 6, 1059 IB_RNR_TIMER_000_12 = 7, 1060 IB_RNR_TIMER_000_16 = 8, 1061 IB_RNR_TIMER_000_24 = 9, 1062 IB_RNR_TIMER_000_32 = 10, 1063 IB_RNR_TIMER_000_48 = 11, 1064 IB_RNR_TIMER_000_64 = 12, 1065 IB_RNR_TIMER_000_96 = 13, 1066 IB_RNR_TIMER_001_28 = 14, 1067 IB_RNR_TIMER_001_92 = 15, 1068 IB_RNR_TIMER_002_56 = 16, 1069 IB_RNR_TIMER_003_84 = 17, 1070 IB_RNR_TIMER_005_12 = 18, 1071 IB_RNR_TIMER_007_68 = 19, 1072 IB_RNR_TIMER_010_24 = 20, 1073 IB_RNR_TIMER_015_36 = 21, 1074 IB_RNR_TIMER_020_48 = 22, 1075 IB_RNR_TIMER_030_72 = 23, 1076 IB_RNR_TIMER_040_96 = 24, 1077 IB_RNR_TIMER_061_44 = 25, 1078 IB_RNR_TIMER_081_92 = 26, 1079 IB_RNR_TIMER_122_88 = 27, 1080 IB_RNR_TIMER_163_84 = 28, 1081 IB_RNR_TIMER_245_76 = 29, 1082 IB_RNR_TIMER_327_68 = 30, 1083 IB_RNR_TIMER_491_52 = 31 1084}; 1085 1086enum ib_qp_attr_mask { 1087 IB_QP_STATE = 1, 1088 IB_QP_CUR_STATE = (1<<1), 1089 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2), 1090 IB_QP_ACCESS_FLAGS = (1<<3), 1091 IB_QP_PKEY_INDEX = (1<<4), 1092 IB_QP_PORT = (1<<5), 1093 IB_QP_QKEY = (1<<6), 1094 IB_QP_AV = (1<<7), 1095 IB_QP_PATH_MTU = (1<<8), 1096 IB_QP_TIMEOUT = (1<<9), 1097 IB_QP_RETRY_CNT = (1<<10), 1098 IB_QP_RNR_RETRY = (1<<11), 1099 IB_QP_RQ_PSN = (1<<12), 1100 IB_QP_MAX_QP_RD_ATOMIC = (1<<13), 1101 IB_QP_ALT_PATH = (1<<14), 1102 IB_QP_MIN_RNR_TIMER = (1<<15), 1103 IB_QP_SQ_PSN = (1<<16), 1104 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17), 1105 IB_QP_PATH_MIG_STATE = (1<<18), 1106 IB_QP_CAP = (1<<19), 1107 IB_QP_DEST_QPN = (1<<20), 1108 IB_QP_RESERVED1 = (1<<21), 1109 IB_QP_RESERVED2 = (1<<22), 1110 IB_QP_RESERVED3 = (1<<23), 1111 IB_QP_RESERVED4 = (1<<24), 1112}; 1113 1114enum ib_qp_state { 1115 IB_QPS_RESET, 1116 IB_QPS_INIT, 1117 IB_QPS_RTR, 1118 IB_QPS_RTS, 1119 IB_QPS_SQD, 1120 IB_QPS_SQE, 1121 IB_QPS_ERR, 1122 IB_QPS_DUMMY = -1, /* force enum signed */ 1123}; 1124 1125enum ib_mig_state { 1126 IB_MIG_MIGRATED, 1127 IB_MIG_REARM, 1128 IB_MIG_ARMED 1129}; 1130 1131enum ib_mw_type { 1132 IB_MW_TYPE_1 = 1, 1133 IB_MW_TYPE_2 = 2 1134}; 1135 1136struct ib_qp_attr { 1137 enum ib_qp_state qp_state; 1138 enum ib_qp_state cur_qp_state; 1139 enum ib_mtu path_mtu; 1140 enum ib_mig_state path_mig_state; 1141 u32 qkey; 1142 u32 rq_psn; 1143 u32 sq_psn; 1144 u32 dest_qp_num; 1145 int qp_access_flags; 1146 struct ib_qp_cap cap; 1147 struct ib_ah_attr ah_attr; 1148 struct ib_ah_attr alt_ah_attr; 1149 u16 pkey_index; 1150 u16 alt_pkey_index; 1151 u8 en_sqd_async_notify; 1152 u8 sq_draining; 1153 u8 max_rd_atomic; 1154 u8 max_dest_rd_atomic; 1155 u8 min_rnr_timer; 1156 u8 port_num; 1157 u8 timeout; 1158 u8 retry_cnt; 1159 u8 rnr_retry; 1160 u8 alt_port_num; 1161 u8 alt_timeout; 1162}; 1163 1164enum ib_wr_opcode { 1165 IB_WR_RDMA_WRITE, 1166 IB_WR_RDMA_WRITE_WITH_IMM, 1167 IB_WR_SEND, 1168 IB_WR_SEND_WITH_IMM, 1169 IB_WR_RDMA_READ, 1170 IB_WR_ATOMIC_CMP_AND_SWP, 1171 IB_WR_ATOMIC_FETCH_AND_ADD, 1172 IB_WR_LSO, 1173 IB_WR_SEND_WITH_INV, 1174 IB_WR_RDMA_READ_WITH_INV, 1175 IB_WR_LOCAL_INV, 1176 IB_WR_REG_MR, 1177 IB_WR_MASKED_ATOMIC_CMP_AND_SWP, 1178 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD, 1179 IB_WR_REG_SIG_MR, 1180 /* reserve values for low level drivers' internal use. 1181 * These values will not be used at all in the ib core layer. 1182 */ 1183 IB_WR_RESERVED1 = 0xf0, 1184 IB_WR_RESERVED2, 1185 IB_WR_RESERVED3, 1186 IB_WR_RESERVED4, 1187 IB_WR_RESERVED5, 1188 IB_WR_RESERVED6, 1189 IB_WR_RESERVED7, 1190 IB_WR_RESERVED8, 1191 IB_WR_RESERVED9, 1192 IB_WR_RESERVED10, 1193 IB_WR_DUMMY = -1, /* force enum signed */ 1194}; 1195 1196enum ib_send_flags { 1197 IB_SEND_FENCE = 1, 1198 IB_SEND_SIGNALED = (1<<1), 1199 IB_SEND_SOLICITED = (1<<2), 1200 IB_SEND_INLINE = (1<<3), 1201 IB_SEND_IP_CSUM = (1<<4), 1202 1203 /* reserve bits 26-31 for low level drivers' internal use */ 1204 IB_SEND_RESERVED_START = (1 << 26), 1205 IB_SEND_RESERVED_END = (1 << 31), 1206}; 1207 1208struct ib_sge { 1209 u64 addr; 1210 u32 length; 1211 u32 lkey; 1212}; 1213 1214struct ib_cqe { 1215 void (*done)(struct ib_cq *cq, struct ib_wc *wc); 1216}; 1217 1218struct ib_send_wr { 1219 struct ib_send_wr *next; 1220 union { 1221 u64 wr_id; 1222 struct ib_cqe *wr_cqe; 1223 }; 1224 struct ib_sge *sg_list; 1225 int num_sge; 1226 enum ib_wr_opcode opcode; 1227 int send_flags; 1228 union { 1229 __be32 imm_data; 1230 u32 invalidate_rkey; 1231 } ex; 1232}; 1233 1234struct ib_rdma_wr { 1235 struct ib_send_wr wr; 1236 u64 remote_addr; 1237 u32 rkey; 1238}; 1239 1240static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr) 1241{ 1242 return container_of(wr, struct ib_rdma_wr, wr); 1243} 1244 1245struct ib_atomic_wr { 1246 struct ib_send_wr wr; 1247 u64 remote_addr; 1248 u64 compare_add; 1249 u64 swap; 1250 u64 compare_add_mask; 1251 u64 swap_mask; 1252 u32 rkey; 1253}; 1254 1255static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr) 1256{ 1257 return container_of(wr, struct ib_atomic_wr, wr); 1258} 1259 1260struct ib_ud_wr { 1261 struct ib_send_wr wr; 1262 struct ib_ah *ah; 1263 void *header; 1264 int hlen; 1265 int mss; 1266 u32 remote_qpn; 1267 u32 remote_qkey; 1268 u16 pkey_index; /* valid for GSI only */ 1269 u8 port_num; /* valid for DR SMPs on switch only */ 1270}; 1271 1272static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr) 1273{ 1274 return container_of(wr, struct ib_ud_wr, wr); 1275} 1276 1277struct ib_reg_wr { 1278 struct ib_send_wr wr; 1279 struct ib_mr *mr; 1280 u32 key; 1281 int access; 1282}; 1283 1284static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr) 1285{ 1286 return container_of(wr, struct ib_reg_wr, wr); 1287} 1288 1289struct ib_sig_handover_wr { 1290 struct ib_send_wr wr; 1291 struct ib_sig_attrs *sig_attrs; 1292 struct ib_mr *sig_mr; 1293 int access_flags; 1294 struct ib_sge *prot; 1295}; 1296 1297static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr) 1298{ 1299 return container_of(wr, struct ib_sig_handover_wr, wr); 1300} 1301 1302struct ib_recv_wr { 1303 struct ib_recv_wr *next; 1304 union { 1305 u64 wr_id; 1306 struct ib_cqe *wr_cqe; 1307 }; 1308 struct ib_sge *sg_list; 1309 int num_sge; 1310}; 1311 1312enum ib_access_flags { 1313 IB_ACCESS_LOCAL_WRITE = 1, 1314 IB_ACCESS_REMOTE_WRITE = (1<<1), 1315 IB_ACCESS_REMOTE_READ = (1<<2), 1316 IB_ACCESS_REMOTE_ATOMIC = (1<<3), 1317 IB_ACCESS_MW_BIND = (1<<4), 1318 IB_ZERO_BASED = (1<<5), 1319 IB_ACCESS_ON_DEMAND = (1<<6), 1320}; 1321 1322struct ib_phys_buf { 1323 u64 addr; 1324 u64 size; 1325}; 1326 1327/* 1328 * XXX: these are apparently used for ->rereg_user_mr, no idea why they 1329 * are hidden here instead of a uapi header! 1330 */ 1331enum ib_mr_rereg_flags { 1332 IB_MR_REREG_TRANS = 1, 1333 IB_MR_REREG_PD = (1<<1), 1334 IB_MR_REREG_ACCESS = (1<<2), 1335 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1) 1336}; 1337 1338struct ib_fmr_attr { 1339 int max_pages; 1340 int max_maps; 1341 u8 page_shift; 1342}; 1343 1344struct ib_umem; 1345 1346struct ib_ucontext { 1347 struct ib_device *device; 1348 struct list_head pd_list; 1349 struct list_head mr_list; 1350 struct list_head mw_list; 1351 struct list_head cq_list; 1352 struct list_head qp_list; 1353 struct list_head srq_list; 1354 struct list_head ah_list; 1355 struct list_head xrcd_list; 1356 struct list_head rule_list; 1357 struct list_head wq_list; 1358 struct list_head rwq_ind_tbl_list; 1359 int closing; 1360 1361 pid_t tgid; 1362#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING 1363 struct rb_root umem_tree; 1364 /* 1365 * Protects .umem_rbroot and tree, as well as odp_mrs_count and 1366 * mmu notifiers registration. 1367 */ 1368 struct rw_semaphore umem_rwsem; 1369 void (*invalidate_range)(struct ib_umem *umem, 1370 unsigned long start, unsigned long end); 1371 1372 struct mmu_notifier mn; 1373 atomic_t notifier_count; 1374 /* A list of umems that don't have private mmu notifier counters yet. */ 1375 struct list_head no_private_counters; 1376 int odp_mrs_count; 1377#endif 1378}; 1379 1380struct ib_uobject { 1381 u64 user_handle; /* handle given to us by userspace */ 1382 struct ib_ucontext *context; /* associated user context */ 1383 void *object; /* containing object */ 1384 struct list_head list; /* link to context's list */ 1385 int id; /* index into kernel idr */ 1386 struct kref ref; 1387 struct rw_semaphore mutex; /* protects .live */ 1388 struct rcu_head rcu; /* kfree_rcu() overhead */ 1389 int live; 1390}; 1391 1392struct ib_udata { 1393 const void __user *inbuf; 1394 void __user *outbuf; 1395 size_t inlen; 1396 size_t outlen; 1397}; 1398 1399struct ib_pd { 1400 u32 local_dma_lkey; 1401 u32 flags; 1402 struct ib_device *device; 1403 struct ib_uobject *uobject; 1404 atomic_t usecnt; /* count all resources */ 1405 1406 u32 unsafe_global_rkey; 1407 1408 /* 1409 * Implementation details of the RDMA core, don't use in drivers: 1410 */ 1411 struct ib_mr *__internal_mr; 1412}; 1413 1414struct ib_xrcd { 1415 struct ib_device *device; 1416 atomic_t usecnt; /* count all exposed resources */ 1417 struct inode *inode; 1418 1419 struct mutex tgt_qp_mutex; 1420 struct list_head tgt_qp_list; 1421}; 1422 1423struct ib_ah { 1424 struct ib_device *device; 1425 struct ib_pd *pd; 1426 struct ib_uobject *uobject; 1427}; 1428 1429typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context); 1430 1431enum ib_poll_context { 1432 IB_POLL_DIRECT, /* caller context, no hw completions */ 1433 IB_POLL_SOFTIRQ, /* poll from softirq context */ 1434 IB_POLL_WORKQUEUE, /* poll from workqueue */ 1435}; 1436 1437struct ib_cq { 1438 struct ib_device *device; 1439 struct ib_uobject *uobject; 1440 ib_comp_handler comp_handler; 1441 void (*event_handler)(struct ib_event *, void *); 1442 void *cq_context; 1443 int cqe; 1444 atomic_t usecnt; /* count number of work queues */ 1445 enum ib_poll_context poll_ctx; 1446 struct work_struct work; 1447}; 1448 1449struct ib_srq { 1450 struct ib_device *device; 1451 struct ib_pd *pd; 1452 struct ib_uobject *uobject; 1453 void (*event_handler)(struct ib_event *, void *); 1454 void *srq_context; 1455 enum ib_srq_type srq_type; 1456 atomic_t usecnt; 1457 1458 union { 1459 struct { 1460 struct ib_xrcd *xrcd; 1461 struct ib_cq *cq; 1462 u32 srq_num; 1463 } xrc; 1464 } ext; 1465}; 1466 1467enum ib_wq_type { 1468 IB_WQT_RQ 1469}; 1470 1471enum ib_wq_state { 1472 IB_WQS_RESET, 1473 IB_WQS_RDY, 1474 IB_WQS_ERR 1475}; 1476 1477struct ib_wq { 1478 struct ib_device *device; 1479 struct ib_uobject *uobject; 1480 void *wq_context; 1481 void (*event_handler)(struct ib_event *, void *); 1482 struct ib_pd *pd; 1483 struct ib_cq *cq; 1484 u32 wq_num; 1485 enum ib_wq_state state; 1486 enum ib_wq_type wq_type; 1487 atomic_t usecnt; 1488}; 1489 1490struct ib_wq_init_attr { 1491 void *wq_context; 1492 enum ib_wq_type wq_type; 1493 u32 max_wr; 1494 u32 max_sge; 1495 struct ib_cq *cq; 1496 void (*event_handler)(struct ib_event *, void *); 1497}; 1498 1499enum ib_wq_attr_mask { 1500 IB_WQ_STATE = 1 << 0, 1501 IB_WQ_CUR_STATE = 1 << 1, 1502}; 1503 1504struct ib_wq_attr { 1505 enum ib_wq_state wq_state; 1506 enum ib_wq_state curr_wq_state; 1507}; 1508 1509struct ib_rwq_ind_table { 1510 struct ib_device *device; 1511 struct ib_uobject *uobject; 1512 atomic_t usecnt; 1513 u32 ind_tbl_num; 1514 u32 log_ind_tbl_size; 1515 struct ib_wq **ind_tbl; 1516}; 1517 1518struct ib_rwq_ind_table_init_attr { 1519 u32 log_ind_tbl_size; 1520 /* Each entry is a pointer to Receive Work Queue */ 1521 struct ib_wq **ind_tbl; 1522}; 1523 1524/* 1525 * @max_write_sge: Maximum SGE elements per RDMA WRITE request. 1526 * @max_read_sge: Maximum SGE elements per RDMA READ request. 1527 */ 1528struct ib_qp { 1529 struct ib_device *device; 1530 struct ib_pd *pd; 1531 struct ib_cq *send_cq; 1532 struct ib_cq *recv_cq; 1533 spinlock_t mr_lock; 1534 struct ib_srq *srq; 1535 struct ib_xrcd *xrcd; /* XRC TGT QPs only */ 1536 struct list_head xrcd_list; 1537 1538 /* count times opened, mcast attaches, flow attaches */ 1539 atomic_t usecnt; 1540 struct list_head open_list; 1541 struct ib_qp *real_qp; 1542 struct ib_uobject *uobject; 1543 void (*event_handler)(struct ib_event *, void *); 1544 void *qp_context; 1545 u32 qp_num; 1546 u32 max_write_sge; 1547 u32 max_read_sge; 1548 enum ib_qp_type qp_type; 1549 struct ib_rwq_ind_table *rwq_ind_tbl; 1550}; 1551 1552struct ib_mr { 1553 struct ib_device *device; 1554 struct ib_pd *pd; 1555 u32 lkey; 1556 u32 rkey; 1557 u64 iova; 1558 u32 length; 1559 unsigned int page_size; 1560 bool need_inval; 1561 union { 1562 struct ib_uobject *uobject; /* user */ 1563 struct list_head qp_entry; /* FR */ 1564 }; 1565}; 1566 1567struct ib_mw { 1568 struct ib_device *device; 1569 struct ib_pd *pd; 1570 struct ib_uobject *uobject; 1571 u32 rkey; 1572 enum ib_mw_type type; 1573}; 1574 1575struct ib_fmr { 1576 struct ib_device *device; 1577 struct ib_pd *pd; 1578 struct list_head list; 1579 u32 lkey; 1580 u32 rkey; 1581}; 1582 1583/* Supported steering options */ 1584enum ib_flow_attr_type { 1585 /* steering according to rule specifications */ 1586 IB_FLOW_ATTR_NORMAL = 0x0, 1587 /* default unicast and multicast rule - 1588 * receive all Eth traffic which isn't steered to any QP 1589 */ 1590 IB_FLOW_ATTR_ALL_DEFAULT = 0x1, 1591 /* default multicast rule - 1592 * receive all Eth multicast traffic which isn't steered to any QP 1593 */ 1594 IB_FLOW_ATTR_MC_DEFAULT = 0x2, 1595 /* sniffer rule - receive all port traffic */ 1596 IB_FLOW_ATTR_SNIFFER = 0x3 1597}; 1598 1599/* Supported steering header types */ 1600enum ib_flow_spec_type { 1601 /* L2 headers*/ 1602 IB_FLOW_SPEC_ETH = 0x20, 1603 IB_FLOW_SPEC_IB = 0x22, 1604 /* L3 header*/ 1605 IB_FLOW_SPEC_IPV4 = 0x30, 1606 IB_FLOW_SPEC_IPV6 = 0x31, 1607 /* L4 headers*/ 1608 IB_FLOW_SPEC_TCP = 0x40, 1609 IB_FLOW_SPEC_UDP = 0x41 1610}; 1611#define IB_FLOW_SPEC_LAYER_MASK 0xF0 1612#define IB_FLOW_SPEC_SUPPORT_LAYERS 4 1613 1614/* Flow steering rule priority is set according to it's domain. 1615 * Lower domain value means higher priority. 1616 */ 1617enum ib_flow_domain { 1618 IB_FLOW_DOMAIN_USER, 1619 IB_FLOW_DOMAIN_ETHTOOL, 1620 IB_FLOW_DOMAIN_RFS, 1621 IB_FLOW_DOMAIN_NIC, 1622 IB_FLOW_DOMAIN_NUM /* Must be last */ 1623}; 1624 1625enum ib_flow_flags { 1626 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */ 1627 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 2 /* Must be last */ 1628}; 1629 1630struct ib_flow_eth_filter { 1631 u8 dst_mac[6]; 1632 u8 src_mac[6]; 1633 __be16 ether_type; 1634 __be16 vlan_tag; 1635 /* Must be last */ 1636 u8 real_sz[0]; 1637}; 1638 1639struct ib_flow_spec_eth { 1640 enum ib_flow_spec_type type; 1641 u16 size; 1642 struct ib_flow_eth_filter val; 1643 struct ib_flow_eth_filter mask; 1644}; 1645 1646struct ib_flow_ib_filter { 1647 __be16 dlid; 1648 __u8 sl; 1649 /* Must be last */ 1650 u8 real_sz[0]; 1651}; 1652 1653struct ib_flow_spec_ib { 1654 enum ib_flow_spec_type type; 1655 u16 size; 1656 struct ib_flow_ib_filter val; 1657 struct ib_flow_ib_filter mask; 1658}; 1659 1660/* IPv4 header flags */ 1661enum ib_ipv4_flags { 1662 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */ 1663 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the 1664 last have this flag set */ 1665}; 1666 1667struct ib_flow_ipv4_filter { 1668 __be32 src_ip; 1669 __be32 dst_ip; 1670 u8 proto; 1671 u8 tos; 1672 u8 ttl; 1673 u8 flags; 1674 /* Must be last */ 1675 u8 real_sz[0]; 1676}; 1677 1678struct ib_flow_spec_ipv4 { 1679 enum ib_flow_spec_type type; 1680 u16 size; 1681 struct ib_flow_ipv4_filter val; 1682 struct ib_flow_ipv4_filter mask; 1683}; 1684 1685struct ib_flow_ipv6_filter { 1686 u8 src_ip[16]; 1687 u8 dst_ip[16]; 1688 __be32 flow_label; 1689 u8 next_hdr; 1690 u8 traffic_class; 1691 u8 hop_limit; 1692 /* Must be last */ 1693 u8 real_sz[0]; 1694}; 1695 1696struct ib_flow_spec_ipv6 { 1697 enum ib_flow_spec_type type; 1698 u16 size; 1699 struct ib_flow_ipv6_filter val; 1700 struct ib_flow_ipv6_filter mask; 1701}; 1702 1703struct ib_flow_tcp_udp_filter { 1704 __be16 dst_port; 1705 __be16 src_port; 1706 /* Must be last */ 1707 u8 real_sz[0]; 1708}; 1709 1710struct ib_flow_spec_tcp_udp { 1711 enum ib_flow_spec_type type; 1712 u16 size; 1713 struct ib_flow_tcp_udp_filter val; 1714 struct ib_flow_tcp_udp_filter mask; 1715}; 1716 1717union ib_flow_spec { 1718 struct { 1719 enum ib_flow_spec_type type; 1720 u16 size; 1721 }; 1722 struct ib_flow_spec_eth eth; 1723 struct ib_flow_spec_ib ib; 1724 struct ib_flow_spec_ipv4 ipv4; 1725 struct ib_flow_spec_tcp_udp tcp_udp; 1726 struct ib_flow_spec_ipv6 ipv6; 1727}; 1728 1729struct ib_flow_attr { 1730 enum ib_flow_attr_type type; 1731 u16 size; 1732 u16 priority; 1733 u32 flags; 1734 u8 num_of_specs; 1735 u8 port; 1736 /* Following are the optional layers according to user request 1737 * struct ib_flow_spec_xxx 1738 * struct ib_flow_spec_yyy 1739 */ 1740}; 1741 1742struct ib_flow { 1743 struct ib_qp *qp; 1744 struct ib_uobject *uobject; 1745}; 1746 1747struct ib_mad_hdr; 1748struct ib_grh; 1749 1750enum ib_process_mad_flags { 1751 IB_MAD_IGNORE_MKEY = 1, 1752 IB_MAD_IGNORE_BKEY = 2, 1753 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY 1754}; 1755 1756enum ib_mad_result { 1757 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */ 1758 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */ 1759 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */ 1760 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */ 1761}; 1762 1763#define IB_DEVICE_NAME_MAX 64 1764 1765struct ib_cache { 1766 rwlock_t lock; 1767 struct ib_event_handler event_handler; 1768 struct ib_pkey_cache **pkey_cache; 1769 struct ib_gid_table **gid_cache; 1770 u8 *lmc_cache; 1771}; 1772 1773struct ib_dma_mapping_ops { 1774 int (*mapping_error)(struct ib_device *dev, 1775 u64 dma_addr); 1776 u64 (*map_single)(struct ib_device *dev, 1777 void *ptr, size_t size, 1778 enum dma_data_direction direction); 1779 void (*unmap_single)(struct ib_device *dev, 1780 u64 addr, size_t size, 1781 enum dma_data_direction direction); 1782 u64 (*map_page)(struct ib_device *dev, 1783 struct page *page, unsigned long offset, 1784 size_t size, 1785 enum dma_data_direction direction); 1786 void (*unmap_page)(struct ib_device *dev, 1787 u64 addr, size_t size, 1788 enum dma_data_direction direction); 1789 int (*map_sg)(struct ib_device *dev, 1790 struct scatterlist *sg, int nents, 1791 enum dma_data_direction direction); 1792 void (*unmap_sg)(struct ib_device *dev, 1793 struct scatterlist *sg, int nents, 1794 enum dma_data_direction direction); 1795 int (*map_sg_attrs)(struct ib_device *dev, 1796 struct scatterlist *sg, int nents, 1797 enum dma_data_direction direction, 1798 struct dma_attrs *attrs); 1799 void (*unmap_sg_attrs)(struct ib_device *dev, 1800 struct scatterlist *sg, int nents, 1801 enum dma_data_direction direction, 1802 struct dma_attrs *attrs); 1803 void (*sync_single_for_cpu)(struct ib_device *dev, 1804 u64 dma_handle, 1805 size_t size, 1806 enum dma_data_direction dir); 1807 void (*sync_single_for_device)(struct ib_device *dev, 1808 u64 dma_handle, 1809 size_t size, 1810 enum dma_data_direction dir); 1811 void *(*alloc_coherent)(struct ib_device *dev, 1812 size_t size, 1813 u64 *dma_handle, 1814 gfp_t flag); 1815 void (*free_coherent)(struct ib_device *dev, 1816 size_t size, void *cpu_addr, 1817 u64 dma_handle); 1818}; 1819 1820struct iw_cm_verbs; 1821 1822struct ib_port_immutable { 1823 int pkey_tbl_len; 1824 int gid_tbl_len; 1825 u32 core_cap_flags; 1826 u32 max_mad_size; 1827}; 1828 1829struct ib_device { 1830 struct device *dma_device; 1831 1832 char name[IB_DEVICE_NAME_MAX]; 1833 1834 struct list_head event_handler_list; 1835 spinlock_t event_handler_lock; 1836 1837 spinlock_t client_data_lock; 1838 struct list_head core_list; 1839 /* Access to the client_data_list is protected by the client_data_lock 1840 * spinlock and the lists_rwsem read-write semaphore */ 1841 struct list_head client_data_list; 1842 1843 struct ib_cache cache; 1844 /** 1845 * port_immutable is indexed by port number 1846 */ 1847 struct ib_port_immutable *port_immutable; 1848 1849 int num_comp_vectors; 1850 1851 struct iw_cm_verbs *iwcm; 1852 1853 /** 1854 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the 1855 * driver initialized data. The struct is kfree()'ed by the sysfs 1856 * core when the device is removed. A lifespan of -1 in the return 1857 * struct tells the core to set a default lifespan. 1858 */ 1859 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device, 1860 u8 port_num); 1861 /** 1862 * get_hw_stats - Fill in the counter value(s) in the stats struct. 1863 * @index - The index in the value array we wish to have updated, or 1864 * num_counters if we want all stats updated 1865 * Return codes - 1866 * < 0 - Error, no counters updated 1867 * index - Updated the single counter pointed to by index 1868 * num_counters - Updated all counters (will reset the timestamp 1869 * and prevent further calls for lifespan milliseconds) 1870 * Drivers are allowed to update all counters in leiu of just the 1871 * one given in index at their option 1872 */ 1873 int (*get_hw_stats)(struct ib_device *device, 1874 struct rdma_hw_stats *stats, 1875 u8 port, int index); 1876 int (*query_device)(struct ib_device *device, 1877 struct ib_device_attr *device_attr, 1878 struct ib_udata *udata); 1879 int (*query_port)(struct ib_device *device, 1880 u8 port_num, 1881 struct ib_port_attr *port_attr); 1882 enum rdma_link_layer (*get_link_layer)(struct ib_device *device, 1883 u8 port_num); 1884 /* When calling get_netdev, the HW vendor's driver should return the 1885 * net device of device @device at port @port_num or NULL if such 1886 * a net device doesn't exist. The vendor driver should call dev_hold 1887 * on this net device. The HW vendor's device driver must guarantee 1888 * that this function returns NULL before the net device reaches 1889 * NETDEV_UNREGISTER_FINAL state. 1890 */ 1891 struct net_device *(*get_netdev)(struct ib_device *device, 1892 u8 port_num); 1893 int (*query_gid)(struct ib_device *device, 1894 u8 port_num, int index, 1895 union ib_gid *gid); 1896 /* When calling add_gid, the HW vendor's driver should 1897 * add the gid of device @device at gid index @index of 1898 * port @port_num to be @gid. Meta-info of that gid (for example, 1899 * the network device related to this gid is available 1900 * at @attr. @context allows the HW vendor driver to store extra 1901 * information together with a GID entry. The HW vendor may allocate 1902 * memory to contain this information and store it in @context when a 1903 * new GID entry is written to. Params are consistent until the next 1904 * call of add_gid or delete_gid. The function should return 0 on 1905 * success or error otherwise. The function could be called 1906 * concurrently for different ports. This function is only called 1907 * when roce_gid_table is used. 1908 */ 1909 int (*add_gid)(struct ib_device *device, 1910 u8 port_num, 1911 unsigned int index, 1912 const union ib_gid *gid, 1913 const struct ib_gid_attr *attr, 1914 void **context); 1915 /* When calling del_gid, the HW vendor's driver should delete the 1916 * gid of device @device at gid index @index of port @port_num. 1917 * Upon the deletion of a GID entry, the HW vendor must free any 1918 * allocated memory. The caller will clear @context afterwards. 1919 * This function is only called when roce_gid_table is used. 1920 */ 1921 int (*del_gid)(struct ib_device *device, 1922 u8 port_num, 1923 unsigned int index, 1924 void **context); 1925 int (*query_pkey)(struct ib_device *device, 1926 u8 port_num, u16 index, u16 *pkey); 1927 int (*modify_device)(struct ib_device *device, 1928 int device_modify_mask, 1929 struct ib_device_modify *device_modify); 1930 int (*modify_port)(struct ib_device *device, 1931 u8 port_num, int port_modify_mask, 1932 struct ib_port_modify *port_modify); 1933 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device, 1934 struct ib_udata *udata); 1935 int (*dealloc_ucontext)(struct ib_ucontext *context); 1936 int (*mmap)(struct ib_ucontext *context, 1937 struct vm_area_struct *vma); 1938 struct ib_pd * (*alloc_pd)(struct ib_device *device, 1939 struct ib_ucontext *context, 1940 struct ib_udata *udata); 1941 int (*dealloc_pd)(struct ib_pd *pd); 1942 struct ib_ah * (*create_ah)(struct ib_pd *pd, 1943 struct ib_ah_attr *ah_attr, 1944 struct ib_udata *udata); 1945 int (*modify_ah)(struct ib_ah *ah, 1946 struct ib_ah_attr *ah_attr); 1947 int (*query_ah)(struct ib_ah *ah, 1948 struct ib_ah_attr *ah_attr); 1949 int (*destroy_ah)(struct ib_ah *ah); 1950 struct ib_srq * (*create_srq)(struct ib_pd *pd, 1951 struct ib_srq_init_attr *srq_init_attr, 1952 struct ib_udata *udata); 1953 int (*modify_srq)(struct ib_srq *srq, 1954 struct ib_srq_attr *srq_attr, 1955 enum ib_srq_attr_mask srq_attr_mask, 1956 struct ib_udata *udata); 1957 int (*query_srq)(struct ib_srq *srq, 1958 struct ib_srq_attr *srq_attr); 1959 int (*destroy_srq)(struct ib_srq *srq); 1960 int (*post_srq_recv)(struct ib_srq *srq, 1961 struct ib_recv_wr *recv_wr, 1962 struct ib_recv_wr **bad_recv_wr); 1963 struct ib_qp * (*create_qp)(struct ib_pd *pd, 1964 struct ib_qp_init_attr *qp_init_attr, 1965 struct ib_udata *udata); 1966 int (*modify_qp)(struct ib_qp *qp, 1967 struct ib_qp_attr *qp_attr, 1968 int qp_attr_mask, 1969 struct ib_udata *udata); 1970 int (*query_qp)(struct ib_qp *qp, 1971 struct ib_qp_attr *qp_attr, 1972 int qp_attr_mask, 1973 struct ib_qp_init_attr *qp_init_attr); 1974 int (*destroy_qp)(struct ib_qp *qp); 1975 int (*post_send)(struct ib_qp *qp, 1976 struct ib_send_wr *send_wr, 1977 struct ib_send_wr **bad_send_wr); 1978 int (*post_recv)(struct ib_qp *qp, 1979 struct ib_recv_wr *recv_wr, 1980 struct ib_recv_wr **bad_recv_wr); 1981 struct ib_cq * (*create_cq)(struct ib_device *device, 1982 const struct ib_cq_init_attr *attr, 1983 struct ib_ucontext *context, 1984 struct ib_udata *udata); 1985 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, 1986 u16 cq_period); 1987 int (*destroy_cq)(struct ib_cq *cq); 1988 int (*resize_cq)(struct ib_cq *cq, int cqe, 1989 struct ib_udata *udata); 1990 int (*poll_cq)(struct ib_cq *cq, int num_entries, 1991 struct ib_wc *wc); 1992 int (*peek_cq)(struct ib_cq *cq, int wc_cnt); 1993 int (*req_notify_cq)(struct ib_cq *cq, 1994 enum ib_cq_notify_flags flags); 1995 int (*req_ncomp_notif)(struct ib_cq *cq, 1996 int wc_cnt); 1997 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd, 1998 int mr_access_flags); 1999 struct ib_mr * (*reg_phys_mr)(struct ib_pd *pd, 2000 struct ib_phys_buf *phys_buf_array, 2001 int num_phys_buf, 2002 int mr_access_flags, 2003 u64 *iova_start); 2004 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd, 2005 u64 start, u64 length, 2006 u64 virt_addr, 2007 int mr_access_flags, 2008 struct ib_udata *udata); 2009 int (*rereg_user_mr)(struct ib_mr *mr, 2010 int flags, 2011 u64 start, u64 length, 2012 u64 virt_addr, 2013 int mr_access_flags, 2014 struct ib_pd *pd, 2015 struct ib_udata *udata); 2016 int (*dereg_mr)(struct ib_mr *mr); 2017 struct ib_mr * (*alloc_mr)(struct ib_pd *pd, 2018 enum ib_mr_type mr_type, 2019 u32 max_num_sg); 2020 int (*map_mr_sg)(struct ib_mr *mr, 2021 struct scatterlist *sg, 2022 int sg_nents, 2023 unsigned int *sg_offset); 2024 struct ib_mw * (*alloc_mw)(struct ib_pd *pd, 2025 enum ib_mw_type type, 2026 struct ib_udata *udata); 2027 int (*dealloc_mw)(struct ib_mw *mw); 2028 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd, 2029 int mr_access_flags, 2030 struct ib_fmr_attr *fmr_attr); 2031 int (*map_phys_fmr)(struct ib_fmr *fmr, 2032 u64 *page_list, int list_len, 2033 u64 iova); 2034 int (*unmap_fmr)(struct list_head *fmr_list); 2035 int (*dealloc_fmr)(struct ib_fmr *fmr); 2036 int (*attach_mcast)(struct ib_qp *qp, 2037 union ib_gid *gid, 2038 u16 lid); 2039 int (*detach_mcast)(struct ib_qp *qp, 2040 union ib_gid *gid, 2041 u16 lid); 2042 int (*process_mad)(struct ib_device *device, 2043 int process_mad_flags, 2044 u8 port_num, 2045 const struct ib_wc *in_wc, 2046 const struct ib_grh *in_grh, 2047 const struct ib_mad_hdr *in_mad, 2048 size_t in_mad_size, 2049 struct ib_mad_hdr *out_mad, 2050 size_t *out_mad_size, 2051 u16 *out_mad_pkey_index); 2052 struct ib_xrcd * (*alloc_xrcd)(struct ib_device *device, 2053 struct ib_ucontext *ucontext, 2054 struct ib_udata *udata); 2055 int (*dealloc_xrcd)(struct ib_xrcd *xrcd); 2056 struct ib_flow * (*create_flow)(struct ib_qp *qp, 2057 struct ib_flow_attr 2058 *flow_attr, 2059 int domain); 2060 int (*destroy_flow)(struct ib_flow *flow_id); 2061 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask, 2062 struct ib_mr_status *mr_status); 2063 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext); 2064 void (*drain_rq)(struct ib_qp *qp); 2065 void (*drain_sq)(struct ib_qp *qp); 2066 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port, 2067 int state); 2068 int (*get_vf_config)(struct ib_device *device, int vf, u8 port, 2069 struct ifla_vf_info *ivf); 2070 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port, 2071 struct ifla_vf_stats *stats); 2072 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid, 2073 int type); 2074 struct ib_wq * (*create_wq)(struct ib_pd *pd, 2075 struct ib_wq_init_attr *init_attr, 2076 struct ib_udata *udata); 2077 int (*destroy_wq)(struct ib_wq *wq); 2078 int (*modify_wq)(struct ib_wq *wq, 2079 struct ib_wq_attr *attr, 2080 u32 wq_attr_mask, 2081 struct ib_udata *udata); 2082 struct ib_rwq_ind_table * (*create_rwq_ind_table)(struct ib_device *device, 2083 struct ib_rwq_ind_table_init_attr *init_attr, 2084 struct ib_udata *udata); 2085 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table); 2086 struct ib_dma_mapping_ops *dma_ops; 2087 2088 struct module *owner; 2089 struct device dev; 2090 struct kobject *ports_parent; 2091 struct list_head port_list; 2092 2093 enum { 2094 IB_DEV_UNINITIALIZED, 2095 IB_DEV_REGISTERED, 2096 IB_DEV_UNREGISTERED 2097 } reg_state; 2098 2099 int uverbs_abi_ver; 2100 u64 uverbs_cmd_mask; 2101 u64 uverbs_ex_cmd_mask; 2102 2103 char node_desc[IB_DEVICE_NODE_DESC_MAX]; 2104 __be64 node_guid; 2105 u32 local_dma_lkey; 2106 u16 is_switch:1; 2107 u8 node_type; 2108 u8 phys_port_cnt; 2109 struct ib_device_attr attrs; 2110 struct attribute_group *hw_stats_ag; 2111 struct rdma_hw_stats *hw_stats; 2112 2113 /** 2114 * The following mandatory functions are used only at device 2115 * registration. Keep functions such as these at the end of this 2116 * structure to avoid cache line misses when accessing struct ib_device 2117 * in fast paths. 2118 */ 2119 int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *); 2120 void (*get_dev_fw_str)(struct ib_device *, char *str, size_t str_len); 2121}; 2122 2123struct ib_client { 2124 char *name; 2125 void (*add) (struct ib_device *); 2126 void (*remove)(struct ib_device *, void *client_data); 2127 2128 /* Returns the net_dev belonging to this ib_client and matching the 2129 * given parameters. 2130 * @dev: An RDMA device that the net_dev use for communication. 2131 * @port: A physical port number on the RDMA device. 2132 * @pkey: P_Key that the net_dev uses if applicable. 2133 * @gid: A GID that the net_dev uses to communicate. 2134 * @addr: An IP address the net_dev is configured with. 2135 * @client_data: The device's client data set by ib_set_client_data(). 2136 * 2137 * An ib_client that implements a net_dev on top of RDMA devices 2138 * (such as IP over IB) should implement this callback, allowing the 2139 * rdma_cm module to find the right net_dev for a given request. 2140 * 2141 * The caller is responsible for calling dev_put on the returned 2142 * netdev. */ 2143 struct net_device *(*get_net_dev_by_params)( 2144 struct ib_device *dev, 2145 u8 port, 2146 u16 pkey, 2147 const union ib_gid *gid, 2148 const struct sockaddr *addr, 2149 void *client_data); 2150 struct list_head list; 2151}; 2152 2153struct ib_device *ib_alloc_device(size_t size); 2154void ib_dealloc_device(struct ib_device *device); 2155 2156void ib_get_device_fw_str(struct ib_device *device, char *str, size_t str_len); 2157 2158int ib_register_device(struct ib_device *device, 2159 int (*port_callback)(struct ib_device *, 2160 u8, struct kobject *)); 2161void ib_unregister_device(struct ib_device *device); 2162 2163int ib_register_client (struct ib_client *client); 2164void ib_unregister_client(struct ib_client *client); 2165 2166void *ib_get_client_data(struct ib_device *device, struct ib_client *client); 2167void ib_set_client_data(struct ib_device *device, struct ib_client *client, 2168 void *data); 2169 2170static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len) 2171{ 2172 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0; 2173} 2174 2175static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len) 2176{ 2177 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0; 2178} 2179 2180static inline bool ib_is_udata_cleared(struct ib_udata *udata, 2181 size_t offset, 2182 size_t len) 2183{ 2184 const void __user *p = (const char __user *)udata->inbuf + offset; 2185 bool ret; 2186 u8 *buf; 2187 2188 if (len > USHRT_MAX) 2189 return false; 2190 2191 buf = memdup_user(p, len); 2192 if (IS_ERR(buf)) 2193 return false; 2194 2195 ret = !memchr_inv(buf, 0, len); 2196 kfree(buf); 2197 return ret; 2198} 2199 2200/** 2201 * ib_modify_qp_is_ok - Check that the supplied attribute mask 2202 * contains all required attributes and no attributes not allowed for 2203 * the given QP state transition. 2204 * @cur_state: Current QP state 2205 * @next_state: Next QP state 2206 * @type: QP type 2207 * @mask: Mask of supplied QP attributes 2208 * @ll : link layer of port 2209 * 2210 * This function is a helper function that a low-level driver's 2211 * modify_qp method can use to validate the consumer's input. It 2212 * checks that cur_state and next_state are valid QP states, that a 2213 * transition from cur_state to next_state is allowed by the IB spec, 2214 * and that the attribute mask supplied is allowed for the transition. 2215 */ 2216int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state, 2217 enum ib_qp_type type, enum ib_qp_attr_mask mask, 2218 enum rdma_link_layer ll); 2219 2220int ib_register_event_handler (struct ib_event_handler *event_handler); 2221int ib_unregister_event_handler(struct ib_event_handler *event_handler); 2222void ib_dispatch_event(struct ib_event *event); 2223 2224int ib_query_port(struct ib_device *device, 2225 u8 port_num, struct ib_port_attr *port_attr); 2226 2227enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device, 2228 u8 port_num); 2229 2230/** 2231 * rdma_cap_ib_switch - Check if the device is IB switch 2232 * @device: Device to check 2233 * 2234 * Device driver is responsible for setting is_switch bit on 2235 * in ib_device structure at init time. 2236 * 2237 * Return: true if the device is IB switch. 2238 */ 2239static inline bool rdma_cap_ib_switch(const struct ib_device *device) 2240{ 2241 return device->is_switch; 2242} 2243 2244/** 2245 * rdma_start_port - Return the first valid port number for the device 2246 * specified 2247 * 2248 * @device: Device to be checked 2249 * 2250 * Return start port number 2251 */ 2252static inline u8 rdma_start_port(const struct ib_device *device) 2253{ 2254 return rdma_cap_ib_switch(device) ? 0 : 1; 2255} 2256 2257/** 2258 * rdma_end_port - Return the last valid port number for the device 2259 * specified 2260 * 2261 * @device: Device to be checked 2262 * 2263 * Return last port number 2264 */ 2265static inline u8 rdma_end_port(const struct ib_device *device) 2266{ 2267 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt; 2268} 2269 2270static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num) 2271{ 2272 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB; 2273} 2274 2275static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num) 2276{ 2277 return device->port_immutable[port_num].core_cap_flags & 2278 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP); 2279} 2280 2281static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num) 2282{ 2283 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP; 2284} 2285 2286static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num) 2287{ 2288 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE; 2289} 2290 2291static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num) 2292{ 2293 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP; 2294} 2295 2296static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num) 2297{ 2298 return rdma_protocol_ib(device, port_num) || 2299 rdma_protocol_roce(device, port_num); 2300} 2301 2302/** 2303 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband 2304 * Management Datagrams. 2305 * @device: Device to check 2306 * @port_num: Port number to check 2307 * 2308 * Management Datagrams (MAD) are a required part of the InfiniBand 2309 * specification and are supported on all InfiniBand devices. A slightly 2310 * extended version are also supported on OPA interfaces. 2311 * 2312 * Return: true if the port supports sending/receiving of MAD packets. 2313 */ 2314static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num) 2315{ 2316 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD; 2317} 2318 2319/** 2320 * rdma_cap_opa_mad - Check if the port of device provides support for OPA 2321 * Management Datagrams. 2322 * @device: Device to check 2323 * @port_num: Port number to check 2324 * 2325 * Intel OmniPath devices extend and/or replace the InfiniBand Management 2326 * datagrams with their own versions. These OPA MADs share many but not all of 2327 * the characteristics of InfiniBand MADs. 2328 * 2329 * OPA MADs differ in the following ways: 2330 * 2331 * 1) MADs are variable size up to 2K 2332 * IBTA defined MADs remain fixed at 256 bytes 2333 * 2) OPA SMPs must carry valid PKeys 2334 * 3) OPA SMP packets are a different format 2335 * 2336 * Return: true if the port supports OPA MAD packet formats. 2337 */ 2338static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num) 2339{ 2340 return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD) 2341 == RDMA_CORE_CAP_OPA_MAD; 2342} 2343 2344/** 2345 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband 2346 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI). 2347 * @device: Device to check 2348 * @port_num: Port number to check 2349 * 2350 * Each InfiniBand node is required to provide a Subnet Management Agent 2351 * that the subnet manager can access. Prior to the fabric being fully 2352 * configured by the subnet manager, the SMA is accessed via a well known 2353 * interface called the Subnet Management Interface (SMI). This interface 2354 * uses directed route packets to communicate with the SM to get around the 2355 * chicken and egg problem of the SM needing to know what's on the fabric 2356 * in order to configure the fabric, and needing to configure the fabric in 2357 * order to send packets to the devices on the fabric. These directed 2358 * route packets do not need the fabric fully configured in order to reach 2359 * their destination. The SMI is the only method allowed to send 2360 * directed route packets on an InfiniBand fabric. 2361 * 2362 * Return: true if the port provides an SMI. 2363 */ 2364static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num) 2365{ 2366 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI; 2367} 2368 2369/** 2370 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband 2371 * Communication Manager. 2372 * @device: Device to check 2373 * @port_num: Port number to check 2374 * 2375 * The InfiniBand Communication Manager is one of many pre-defined General 2376 * Service Agents (GSA) that are accessed via the General Service 2377 * Interface (GSI). It's role is to facilitate establishment of connections 2378 * between nodes as well as other management related tasks for established 2379 * connections. 2380 * 2381 * Return: true if the port supports an IB CM (this does not guarantee that 2382 * a CM is actually running however). 2383 */ 2384static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num) 2385{ 2386 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM; 2387} 2388 2389/** 2390 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP 2391 * Communication Manager. 2392 * @device: Device to check 2393 * @port_num: Port number to check 2394 * 2395 * Similar to above, but specific to iWARP connections which have a different 2396 * managment protocol than InfiniBand. 2397 * 2398 * Return: true if the port supports an iWARP CM (this does not guarantee that 2399 * a CM is actually running however). 2400 */ 2401static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num) 2402{ 2403 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM; 2404} 2405 2406/** 2407 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband 2408 * Subnet Administration. 2409 * @device: Device to check 2410 * @port_num: Port number to check 2411 * 2412 * An InfiniBand Subnet Administration (SA) service is a pre-defined General 2413 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand 2414 * fabrics, devices should resolve routes to other hosts by contacting the 2415 * SA to query the proper route. 2416 * 2417 * Return: true if the port should act as a client to the fabric Subnet 2418 * Administration interface. This does not imply that the SA service is 2419 * running locally. 2420 */ 2421static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num) 2422{ 2423 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA; 2424} 2425 2426/** 2427 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband 2428 * Multicast. 2429 * @device: Device to check 2430 * @port_num: Port number to check 2431 * 2432 * InfiniBand multicast registration is more complex than normal IPv4 or 2433 * IPv6 multicast registration. Each Host Channel Adapter must register 2434 * with the Subnet Manager when it wishes to join a multicast group. It 2435 * should do so only once regardless of how many queue pairs it subscribes 2436 * to this group. And it should leave the group only after all queue pairs 2437 * attached to the group have been detached. 2438 * 2439 * Return: true if the port must undertake the additional adminstrative 2440 * overhead of registering/unregistering with the SM and tracking of the 2441 * total number of queue pairs attached to the multicast group. 2442 */ 2443static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num) 2444{ 2445 return rdma_cap_ib_sa(device, port_num); 2446} 2447 2448/** 2449 * rdma_cap_af_ib - Check if the port of device has the capability 2450 * Native Infiniband Address. 2451 * @device: Device to check 2452 * @port_num: Port number to check 2453 * 2454 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default 2455 * GID. RoCE uses a different mechanism, but still generates a GID via 2456 * a prescribed mechanism and port specific data. 2457 * 2458 * Return: true if the port uses a GID address to identify devices on the 2459 * network. 2460 */ 2461static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num) 2462{ 2463 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB; 2464} 2465 2466/** 2467 * rdma_cap_eth_ah - Check if the port of device has the capability 2468 * Ethernet Address Handle. 2469 * @device: Device to check 2470 * @port_num: Port number to check 2471 * 2472 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique 2473 * to fabricate GIDs over Ethernet/IP specific addresses native to the 2474 * port. Normally, packet headers are generated by the sending host 2475 * adapter, but when sending connectionless datagrams, we must manually 2476 * inject the proper headers for the fabric we are communicating over. 2477 * 2478 * Return: true if we are running as a RoCE port and must force the 2479 * addition of a Global Route Header built from our Ethernet Address 2480 * Handle into our header list for connectionless packets. 2481 */ 2482static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num) 2483{ 2484 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH; 2485} 2486 2487/** 2488 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port. 2489 * 2490 * @device: Device 2491 * @port_num: Port number 2492 * 2493 * This MAD size includes the MAD headers and MAD payload. No other headers 2494 * are included. 2495 * 2496 * Return the max MAD size required by the Port. Will return 0 if the port 2497 * does not support MADs 2498 */ 2499static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num) 2500{ 2501 return device->port_immutable[port_num].max_mad_size; 2502} 2503 2504/** 2505 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table 2506 * @device: Device to check 2507 * @port_num: Port number to check 2508 * 2509 * RoCE GID table mechanism manages the various GIDs for a device. 2510 * 2511 * NOTE: if allocating the port's GID table has failed, this call will still 2512 * return true, but any RoCE GID table API will fail. 2513 * 2514 * Return: true if the port uses RoCE GID table mechanism in order to manage 2515 * its GIDs. 2516 */ 2517static inline bool rdma_cap_roce_gid_table(const struct ib_device *device, 2518 u8 port_num) 2519{ 2520 return rdma_protocol_roce(device, port_num) && 2521 device->add_gid && device->del_gid; 2522} 2523 2524/* 2525 * Check if the device supports READ W/ INVALIDATE. 2526 */ 2527static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num) 2528{ 2529 /* 2530 * iWarp drivers must support READ W/ INVALIDATE. No other protocol 2531 * has support for it yet. 2532 */ 2533 return rdma_protocol_iwarp(dev, port_num); 2534} 2535 2536int ib_query_gid(struct ib_device *device, 2537 u8 port_num, int index, union ib_gid *gid, 2538 struct ib_gid_attr *attr); 2539 2540int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port, 2541 int state); 2542int ib_get_vf_config(struct ib_device *device, int vf, u8 port, 2543 struct ifla_vf_info *info); 2544int ib_get_vf_stats(struct ib_device *device, int vf, u8 port, 2545 struct ifla_vf_stats *stats); 2546int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid, 2547 int type); 2548 2549int ib_query_pkey(struct ib_device *device, 2550 u8 port_num, u16 index, u16 *pkey); 2551 2552int ib_modify_device(struct ib_device *device, 2553 int device_modify_mask, 2554 struct ib_device_modify *device_modify); 2555 2556int ib_modify_port(struct ib_device *device, 2557 u8 port_num, int port_modify_mask, 2558 struct ib_port_modify *port_modify); 2559 2560int ib_find_gid(struct ib_device *device, union ib_gid *gid, 2561 enum ib_gid_type gid_type, struct net_device *ndev, 2562 u8 *port_num, u16 *index); 2563 2564int ib_find_pkey(struct ib_device *device, 2565 u8 port_num, u16 pkey, u16 *index); 2566 2567enum ib_pd_flags { 2568 /* 2569 * Create a memory registration for all memory in the system and place 2570 * the rkey for it into pd->unsafe_global_rkey. This can be used by 2571 * ULPs to avoid the overhead of dynamic MRs. 2572 * 2573 * This flag is generally considered unsafe and must only be used in 2574 * extremly trusted environments. Every use of it will log a warning 2575 * in the kernel log. 2576 */ 2577 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01, 2578}; 2579 2580struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags, 2581 const char *caller); 2582#define ib_alloc_pd(device, flags) \ 2583 __ib_alloc_pd((device), (flags), __func__) 2584void ib_dealloc_pd(struct ib_pd *pd); 2585 2586/** 2587 * ib_create_ah - Creates an address handle for the given address vector. 2588 * @pd: The protection domain associated with the address handle. 2589 * @ah_attr: The attributes of the address vector. 2590 * 2591 * The address handle is used to reference a local or global destination 2592 * in all UD QP post sends. 2593 */ 2594struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr); 2595 2596/** 2597 * ib_init_ah_from_wc - Initializes address handle attributes from a 2598 * work completion. 2599 * @device: Device on which the received message arrived. 2600 * @port_num: Port on which the received message arrived. 2601 * @wc: Work completion associated with the received message. 2602 * @grh: References the received global route header. This parameter is 2603 * ignored unless the work completion indicates that the GRH is valid. 2604 * @ah_attr: Returned attributes that can be used when creating an address 2605 * handle for replying to the message. 2606 */ 2607int ib_init_ah_from_wc(struct ib_device *device, u8 port_num, 2608 const struct ib_wc *wc, const struct ib_grh *grh, 2609 struct ib_ah_attr *ah_attr); 2610 2611/** 2612 * ib_create_ah_from_wc - Creates an address handle associated with the 2613 * sender of the specified work completion. 2614 * @pd: The protection domain associated with the address handle. 2615 * @wc: Work completion information associated with a received message. 2616 * @grh: References the received global route header. This parameter is 2617 * ignored unless the work completion indicates that the GRH is valid. 2618 * @port_num: The outbound port number to associate with the address. 2619 * 2620 * The address handle is used to reference a local or global destination 2621 * in all UD QP post sends. 2622 */ 2623struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc, 2624 const struct ib_grh *grh, u8 port_num); 2625 2626/** 2627 * ib_modify_ah - Modifies the address vector associated with an address 2628 * handle. 2629 * @ah: The address handle to modify. 2630 * @ah_attr: The new address vector attributes to associate with the 2631 * address handle. 2632 */ 2633int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr); 2634 2635/** 2636 * ib_query_ah - Queries the address vector associated with an address 2637 * handle. 2638 * @ah: The address handle to query. 2639 * @ah_attr: The address vector attributes associated with the address 2640 * handle. 2641 */ 2642int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr); 2643 2644/** 2645 * ib_destroy_ah - Destroys an address handle. 2646 * @ah: The address handle to destroy. 2647 */ 2648int ib_destroy_ah(struct ib_ah *ah); 2649 2650/** 2651 * ib_create_srq - Creates a SRQ associated with the specified protection 2652 * domain. 2653 * @pd: The protection domain associated with the SRQ. 2654 * @srq_init_attr: A list of initial attributes required to create the 2655 * SRQ. If SRQ creation succeeds, then the attributes are updated to 2656 * the actual capabilities of the created SRQ. 2657 * 2658 * srq_attr->max_wr and srq_attr->max_sge are read the determine the 2659 * requested size of the SRQ, and set to the actual values allocated 2660 * on return. If ib_create_srq() succeeds, then max_wr and max_sge 2661 * will always be at least as large as the requested values. 2662 */ 2663struct ib_srq *ib_create_srq(struct ib_pd *pd, 2664 struct ib_srq_init_attr *srq_init_attr); 2665 2666/** 2667 * ib_modify_srq - Modifies the attributes for the specified SRQ. 2668 * @srq: The SRQ to modify. 2669 * @srq_attr: On input, specifies the SRQ attributes to modify. On output, 2670 * the current values of selected SRQ attributes are returned. 2671 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ 2672 * are being modified. 2673 * 2674 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or 2675 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when 2676 * the number of receives queued drops below the limit. 2677 */ 2678int ib_modify_srq(struct ib_srq *srq, 2679 struct ib_srq_attr *srq_attr, 2680 enum ib_srq_attr_mask srq_attr_mask); 2681 2682/** 2683 * ib_query_srq - Returns the attribute list and current values for the 2684 * specified SRQ. 2685 * @srq: The SRQ to query. 2686 * @srq_attr: The attributes of the specified SRQ. 2687 */ 2688int ib_query_srq(struct ib_srq *srq, 2689 struct ib_srq_attr *srq_attr); 2690 2691/** 2692 * ib_destroy_srq - Destroys the specified SRQ. 2693 * @srq: The SRQ to destroy. 2694 */ 2695int ib_destroy_srq(struct ib_srq *srq); 2696 2697/** 2698 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ. 2699 * @srq: The SRQ to post the work request on. 2700 * @recv_wr: A list of work requests to post on the receive queue. 2701 * @bad_recv_wr: On an immediate failure, this parameter will reference 2702 * the work request that failed to be posted on the QP. 2703 */ 2704static inline int ib_post_srq_recv(struct ib_srq *srq, 2705 struct ib_recv_wr *recv_wr, 2706 struct ib_recv_wr **bad_recv_wr) 2707{ 2708 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr); 2709} 2710 2711/** 2712 * ib_create_qp - Creates a QP associated with the specified protection 2713 * domain. 2714 * @pd: The protection domain associated with the QP. 2715 * @qp_init_attr: A list of initial attributes required to create the 2716 * QP. If QP creation succeeds, then the attributes are updated to 2717 * the actual capabilities of the created QP. 2718 */ 2719struct ib_qp *ib_create_qp(struct ib_pd *pd, 2720 struct ib_qp_init_attr *qp_init_attr); 2721 2722/** 2723 * ib_modify_qp - Modifies the attributes for the specified QP and then 2724 * transitions the QP to the given state. 2725 * @qp: The QP to modify. 2726 * @qp_attr: On input, specifies the QP attributes to modify. On output, 2727 * the current values of selected QP attributes are returned. 2728 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP 2729 * are being modified. 2730 */ 2731int ib_modify_qp(struct ib_qp *qp, 2732 struct ib_qp_attr *qp_attr, 2733 int qp_attr_mask); 2734 2735/** 2736 * ib_query_qp - Returns the attribute list and current values for the 2737 * specified QP. 2738 * @qp: The QP to query. 2739 * @qp_attr: The attributes of the specified QP. 2740 * @qp_attr_mask: A bit-mask used to select specific attributes to query. 2741 * @qp_init_attr: Additional attributes of the selected QP. 2742 * 2743 * The qp_attr_mask may be used to limit the query to gathering only the 2744 * selected attributes. 2745 */ 2746int ib_query_qp(struct ib_qp *qp, 2747 struct ib_qp_attr *qp_attr, 2748 int qp_attr_mask, 2749 struct ib_qp_init_attr *qp_init_attr); 2750 2751/** 2752 * ib_destroy_qp - Destroys the specified QP. 2753 * @qp: The QP to destroy. 2754 */ 2755int ib_destroy_qp(struct ib_qp *qp); 2756 2757/** 2758 * ib_open_qp - Obtain a reference to an existing sharable QP. 2759 * @xrcd - XRC domain 2760 * @qp_open_attr: Attributes identifying the QP to open. 2761 * 2762 * Returns a reference to a sharable QP. 2763 */ 2764struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd, 2765 struct ib_qp_open_attr *qp_open_attr); 2766 2767/** 2768 * ib_close_qp - Release an external reference to a QP. 2769 * @qp: The QP handle to release 2770 * 2771 * The opened QP handle is released by the caller. The underlying 2772 * shared QP is not destroyed until all internal references are released. 2773 */ 2774int ib_close_qp(struct ib_qp *qp); 2775 2776/** 2777 * ib_post_send - Posts a list of work requests to the send queue of 2778 * the specified QP. 2779 * @qp: The QP to post the work request on. 2780 * @send_wr: A list of work requests to post on the send queue. 2781 * @bad_send_wr: On an immediate failure, this parameter will reference 2782 * the work request that failed to be posted on the QP. 2783 * 2784 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate 2785 * error is returned, the QP state shall not be affected, 2786 * ib_post_send() will return an immediate error after queueing any 2787 * earlier work requests in the list. 2788 */ 2789static inline int ib_post_send(struct ib_qp *qp, 2790 struct ib_send_wr *send_wr, 2791 struct ib_send_wr **bad_send_wr) 2792{ 2793 return qp->device->post_send(qp, send_wr, bad_send_wr); 2794} 2795 2796/** 2797 * ib_post_recv - Posts a list of work requests to the receive queue of 2798 * the specified QP. 2799 * @qp: The QP to post the work request on. 2800 * @recv_wr: A list of work requests to post on the receive queue. 2801 * @bad_recv_wr: On an immediate failure, this parameter will reference 2802 * the work request that failed to be posted on the QP. 2803 */ 2804static inline int ib_post_recv(struct ib_qp *qp, 2805 struct ib_recv_wr *recv_wr, 2806 struct ib_recv_wr **bad_recv_wr) 2807{ 2808 return qp->device->post_recv(qp, recv_wr, bad_recv_wr); 2809} 2810 2811struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private, 2812 int nr_cqe, int comp_vector, enum ib_poll_context poll_ctx); 2813void ib_free_cq(struct ib_cq *cq); 2814 2815/** 2816 * ib_create_cq - Creates a CQ on the specified device. 2817 * @device: The device on which to create the CQ. 2818 * @comp_handler: A user-specified callback that is invoked when a 2819 * completion event occurs on the CQ. 2820 * @event_handler: A user-specified callback that is invoked when an 2821 * asynchronous event not associated with a completion occurs on the CQ. 2822 * @cq_context: Context associated with the CQ returned to the user via 2823 * the associated completion and event handlers. 2824 * @cq_attr: The attributes the CQ should be created upon. 2825 * 2826 * Users can examine the cq structure to determine the actual CQ size. 2827 */ 2828struct ib_cq *ib_create_cq(struct ib_device *device, 2829 ib_comp_handler comp_handler, 2830 void (*event_handler)(struct ib_event *, void *), 2831 void *cq_context, 2832 const struct ib_cq_init_attr *cq_attr); 2833 2834/** 2835 * ib_resize_cq - Modifies the capacity of the CQ. 2836 * @cq: The CQ to resize. 2837 * @cqe: The minimum size of the CQ. 2838 * 2839 * Users can examine the cq structure to determine the actual CQ size. 2840 */ 2841int ib_resize_cq(struct ib_cq *cq, int cqe); 2842 2843/** 2844 * ib_modify_cq - Modifies moderation params of the CQ 2845 * @cq: The CQ to modify. 2846 * @cq_count: number of CQEs that will trigger an event 2847 * @cq_period: max period of time in usec before triggering an event 2848 * 2849 */ 2850int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period); 2851 2852/** 2853 * ib_destroy_cq - Destroys the specified CQ. 2854 * @cq: The CQ to destroy. 2855 */ 2856int ib_destroy_cq(struct ib_cq *cq); 2857 2858/** 2859 * ib_poll_cq - poll a CQ for completion(s) 2860 * @cq:the CQ being polled 2861 * @num_entries:maximum number of completions to return 2862 * @wc:array of at least @num_entries &struct ib_wc where completions 2863 * will be returned 2864 * 2865 * Poll a CQ for (possibly multiple) completions. If the return value 2866 * is < 0, an error occurred. If the return value is >= 0, it is the 2867 * number of completions returned. If the return value is 2868 * non-negative and < num_entries, then the CQ was emptied. 2869 */ 2870static inline int ib_poll_cq(struct ib_cq *cq, int num_entries, 2871 struct ib_wc *wc) 2872{ 2873 return cq->device->poll_cq(cq, num_entries, wc); 2874} 2875 2876/** 2877 * ib_peek_cq - Returns the number of unreaped completions currently 2878 * on the specified CQ. 2879 * @cq: The CQ to peek. 2880 * @wc_cnt: A minimum number of unreaped completions to check for. 2881 * 2882 * If the number of unreaped completions is greater than or equal to wc_cnt, 2883 * this function returns wc_cnt, otherwise, it returns the actual number of 2884 * unreaped completions. 2885 */ 2886int ib_peek_cq(struct ib_cq *cq, int wc_cnt); 2887 2888/** 2889 * ib_req_notify_cq - Request completion notification on a CQ. 2890 * @cq: The CQ to generate an event for. 2891 * @flags: 2892 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP 2893 * to request an event on the next solicited event or next work 2894 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS 2895 * may also be |ed in to request a hint about missed events, as 2896 * described below. 2897 * 2898 * Return Value: 2899 * < 0 means an error occurred while requesting notification 2900 * == 0 means notification was requested successfully, and if 2901 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events 2902 * were missed and it is safe to wait for another event. In 2903 * this case is it guaranteed that any work completions added 2904 * to the CQ since the last CQ poll will trigger a completion 2905 * notification event. 2906 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed 2907 * in. It means that the consumer must poll the CQ again to 2908 * make sure it is empty to avoid missing an event because of a 2909 * race between requesting notification and an entry being 2910 * added to the CQ. This return value means it is possible 2911 * (but not guaranteed) that a work completion has been added 2912 * to the CQ since the last poll without triggering a 2913 * completion notification event. 2914 */ 2915static inline int ib_req_notify_cq(struct ib_cq *cq, 2916 enum ib_cq_notify_flags flags) 2917{ 2918 return cq->device->req_notify_cq(cq, flags); 2919} 2920 2921/** 2922 * ib_req_ncomp_notif - Request completion notification when there are 2923 * at least the specified number of unreaped completions on the CQ. 2924 * @cq: The CQ to generate an event for. 2925 * @wc_cnt: The number of unreaped completions that should be on the 2926 * CQ before an event is generated. 2927 */ 2928static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt) 2929{ 2930 return cq->device->req_ncomp_notif ? 2931 cq->device->req_ncomp_notif(cq, wc_cnt) : 2932 -ENOSYS; 2933} 2934 2935/** 2936 * ib_dma_mapping_error - check a DMA addr for error 2937 * @dev: The device for which the dma_addr was created 2938 * @dma_addr: The DMA address to check 2939 */ 2940static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr) 2941{ 2942 if (dev->dma_ops) 2943 return dev->dma_ops->mapping_error(dev, dma_addr); 2944 return dma_mapping_error(dev->dma_device, dma_addr); 2945} 2946 2947/** 2948 * ib_dma_map_single - Map a kernel virtual address to DMA address 2949 * @dev: The device for which the dma_addr is to be created 2950 * @cpu_addr: The kernel virtual address 2951 * @size: The size of the region in bytes 2952 * @direction: The direction of the DMA 2953 */ 2954static inline u64 ib_dma_map_single(struct ib_device *dev, 2955 void *cpu_addr, size_t size, 2956 enum dma_data_direction direction) 2957{ 2958 if (dev->dma_ops) 2959 return dev->dma_ops->map_single(dev, cpu_addr, size, direction); 2960 return dma_map_single(dev->dma_device, cpu_addr, size, direction); 2961} 2962 2963/** 2964 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single() 2965 * @dev: The device for which the DMA address was created 2966 * @addr: The DMA address 2967 * @size: The size of the region in bytes 2968 * @direction: The direction of the DMA 2969 */ 2970static inline void ib_dma_unmap_single(struct ib_device *dev, 2971 u64 addr, size_t size, 2972 enum dma_data_direction direction) 2973{ 2974 if (dev->dma_ops) 2975 dev->dma_ops->unmap_single(dev, addr, size, direction); 2976 else 2977 dma_unmap_single(dev->dma_device, addr, size, direction); 2978} 2979 2980static inline u64 ib_dma_map_single_attrs(struct ib_device *dev, 2981 void *cpu_addr, size_t size, 2982 enum dma_data_direction direction, 2983 struct dma_attrs *dma_attrs) 2984{ 2985 return dma_map_single_attrs(dev->dma_device, cpu_addr, size, 2986 direction, dma_attrs); 2987} 2988 2989static inline void ib_dma_unmap_single_attrs(struct ib_device *dev, 2990 u64 addr, size_t size, 2991 enum dma_data_direction direction, 2992 struct dma_attrs *dma_attrs) 2993{ 2994 return dma_unmap_single_attrs(dev->dma_device, addr, size, 2995 direction, dma_attrs); 2996} 2997 2998/** 2999 * ib_dma_map_page - Map a physical page to DMA address 3000 * @dev: The device for which the dma_addr is to be created 3001 * @page: The page to be mapped 3002 * @offset: The offset within the page 3003 * @size: The size of the region in bytes 3004 * @direction: The direction of the DMA 3005 */ 3006static inline u64 ib_dma_map_page(struct ib_device *dev, 3007 struct page *page, 3008 unsigned long offset, 3009 size_t size, 3010 enum dma_data_direction direction) 3011{ 3012 if (dev->dma_ops) 3013 return dev->dma_ops->map_page(dev, page, offset, size, direction); 3014 return dma_map_page(dev->dma_device, page, offset, size, direction); 3015} 3016 3017/** 3018 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page() 3019 * @dev: The device for which the DMA address was created 3020 * @addr: The DMA address 3021 * @size: The size of the region in bytes 3022 * @direction: The direction of the DMA 3023 */ 3024static inline void ib_dma_unmap_page(struct ib_device *dev, 3025 u64 addr, size_t size, 3026 enum dma_data_direction direction) 3027{ 3028 if (dev->dma_ops) 3029 dev->dma_ops->unmap_page(dev, addr, size, direction); 3030 else 3031 dma_unmap_page(dev->dma_device, addr, size, direction); 3032} 3033 3034/** 3035 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses 3036 * @dev: The device for which the DMA addresses are to be created 3037 * @sg: The array of scatter/gather entries 3038 * @nents: The number of scatter/gather entries 3039 * @direction: The direction of the DMA 3040 */ 3041static inline int ib_dma_map_sg(struct ib_device *dev, 3042 struct scatterlist *sg, int nents, 3043 enum dma_data_direction direction) 3044{ 3045 if (dev->dma_ops) 3046 return dev->dma_ops->map_sg(dev, sg, nents, direction); 3047 return dma_map_sg(dev->dma_device, sg, nents, direction); 3048} 3049 3050/** 3051 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses 3052 * @dev: The device for which the DMA addresses were created 3053 * @sg: The array of scatter/gather entries 3054 * @nents: The number of scatter/gather entries 3055 * @direction: The direction of the DMA 3056 */ 3057static inline void ib_dma_unmap_sg(struct ib_device *dev, 3058 struct scatterlist *sg, int nents, 3059 enum dma_data_direction direction) 3060{ 3061 if (dev->dma_ops) 3062 dev->dma_ops->unmap_sg(dev, sg, nents, direction); 3063 else 3064 dma_unmap_sg(dev->dma_device, sg, nents, direction); 3065} 3066 3067static inline int ib_dma_map_sg_attrs(struct ib_device *dev, 3068 struct scatterlist *sg, int nents, 3069 enum dma_data_direction direction, 3070 struct dma_attrs *dma_attrs) 3071{ 3072 if (dev->dma_ops) 3073 return dev->dma_ops->map_sg_attrs(dev, sg, nents, direction, 3074 dma_attrs); 3075 else 3076 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, 3077 dma_attrs); 3078} 3079 3080static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev, 3081 struct scatterlist *sg, int nents, 3082 enum dma_data_direction direction, 3083 struct dma_attrs *dma_attrs) 3084{ 3085 if (dev->dma_ops) 3086 return dev->dma_ops->unmap_sg_attrs(dev, sg, nents, direction, 3087 dma_attrs); 3088 else 3089 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, 3090 dma_attrs); 3091} 3092/** 3093 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry 3094 * @dev: The device for which the DMA addresses were created 3095 * @sg: The scatter/gather entry 3096 * 3097 * Note: this function is obsolete. To do: change all occurrences of 3098 * ib_sg_dma_address() into sg_dma_address(). 3099 */ 3100static inline u64 ib_sg_dma_address(struct ib_device *dev, 3101 struct scatterlist *sg) 3102{ 3103 return sg_dma_address(sg); 3104} 3105 3106/** 3107 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry 3108 * @dev: The device for which the DMA addresses were created 3109 * @sg: The scatter/gather entry 3110 * 3111 * Note: this function is obsolete. To do: change all occurrences of 3112 * ib_sg_dma_len() into sg_dma_len(). 3113 */ 3114static inline unsigned int ib_sg_dma_len(struct ib_device *dev, 3115 struct scatterlist *sg) 3116{ 3117 return sg_dma_len(sg); 3118} 3119 3120/** 3121 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU 3122 * @dev: The device for which the DMA address was created 3123 * @addr: The DMA address 3124 * @size: The size of the region in bytes 3125 * @dir: The direction of the DMA 3126 */ 3127static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev, 3128 u64 addr, 3129 size_t size, 3130 enum dma_data_direction dir) 3131{ 3132 if (dev->dma_ops) 3133 dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir); 3134 else 3135 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir); 3136} 3137 3138/** 3139 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device 3140 * @dev: The device for which the DMA address was created 3141 * @addr: The DMA address 3142 * @size: The size of the region in bytes 3143 * @dir: The direction of the DMA 3144 */ 3145static inline void ib_dma_sync_single_for_device(struct ib_device *dev, 3146 u64 addr, 3147 size_t size, 3148 enum dma_data_direction dir) 3149{ 3150 if (dev->dma_ops) 3151 dev->dma_ops->sync_single_for_device(dev, addr, size, dir); 3152 else 3153 dma_sync_single_for_device(dev->dma_device, addr, size, dir); 3154} 3155 3156/** 3157 * ib_dma_alloc_coherent - Allocate memory and map it for DMA 3158 * @dev: The device for which the DMA address is requested 3159 * @size: The size of the region to allocate in bytes 3160 * @dma_handle: A pointer for returning the DMA address of the region 3161 * @flag: memory allocator flags 3162 */ 3163static inline void *ib_dma_alloc_coherent(struct ib_device *dev, 3164 size_t size, 3165 u64 *dma_handle, 3166 gfp_t flag) 3167{ 3168 if (dev->dma_ops) 3169 return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag); 3170 else { 3171 dma_addr_t handle; 3172 void *ret; 3173 3174 ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag); 3175 *dma_handle = handle; 3176 return ret; 3177 } 3178} 3179 3180/** 3181 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent() 3182 * @dev: The device for which the DMA addresses were allocated 3183 * @size: The size of the region 3184 * @cpu_addr: the address returned by ib_dma_alloc_coherent() 3185 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent() 3186 */ 3187static inline void ib_dma_free_coherent(struct ib_device *dev, 3188 size_t size, void *cpu_addr, 3189 u64 dma_handle) 3190{ 3191 if (dev->dma_ops) 3192 dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle); 3193 else 3194 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle); 3195} 3196 3197/** 3198 * ib_dereg_mr - Deregisters a memory region and removes it from the 3199 * HCA translation table. 3200 * @mr: The memory region to deregister. 3201 * 3202 * This function can fail, if the memory region has memory windows bound to it. 3203 */ 3204int ib_dereg_mr(struct ib_mr *mr); 3205 3206struct ib_mr *ib_alloc_mr(struct ib_pd *pd, 3207 enum ib_mr_type mr_type, 3208 u32 max_num_sg); 3209 3210/** 3211 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR 3212 * R_Key and L_Key. 3213 * @mr - struct ib_mr pointer to be updated. 3214 * @newkey - new key to be used. 3215 */ 3216static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey) 3217{ 3218 mr->lkey = (mr->lkey & 0xffffff00) | newkey; 3219 mr->rkey = (mr->rkey & 0xffffff00) | newkey; 3220} 3221 3222/** 3223 * ib_inc_rkey - increments the key portion of the given rkey. Can be used 3224 * for calculating a new rkey for type 2 memory windows. 3225 * @rkey - the rkey to increment. 3226 */ 3227static inline u32 ib_inc_rkey(u32 rkey) 3228{ 3229 const u32 mask = 0x000000ff; 3230 return ((rkey + 1) & mask) | (rkey & ~mask); 3231} 3232 3233/** 3234 * ib_alloc_fmr - Allocates a unmapped fast memory region. 3235 * @pd: The protection domain associated with the unmapped region. 3236 * @mr_access_flags: Specifies the memory access rights. 3237 * @fmr_attr: Attributes of the unmapped region. 3238 * 3239 * A fast memory region must be mapped before it can be used as part of 3240 * a work request. 3241 */ 3242struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd, 3243 int mr_access_flags, 3244 struct ib_fmr_attr *fmr_attr); 3245 3246/** 3247 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region. 3248 * @fmr: The fast memory region to associate with the pages. 3249 * @page_list: An array of physical pages to map to the fast memory region. 3250 * @list_len: The number of pages in page_list. 3251 * @iova: The I/O virtual address to use with the mapped region. 3252 */ 3253static inline int ib_map_phys_fmr(struct ib_fmr *fmr, 3254 u64 *page_list, int list_len, 3255 u64 iova) 3256{ 3257 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova); 3258} 3259 3260/** 3261 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions. 3262 * @fmr_list: A linked list of fast memory regions to unmap. 3263 */ 3264int ib_unmap_fmr(struct list_head *fmr_list); 3265 3266/** 3267 * ib_dealloc_fmr - Deallocates a fast memory region. 3268 * @fmr: The fast memory region to deallocate. 3269 */ 3270int ib_dealloc_fmr(struct ib_fmr *fmr); 3271 3272/** 3273 * ib_attach_mcast - Attaches the specified QP to a multicast group. 3274 * @qp: QP to attach to the multicast group. The QP must be type 3275 * IB_QPT_UD. 3276 * @gid: Multicast group GID. 3277 * @lid: Multicast group LID in host byte order. 3278 * 3279 * In order to send and receive multicast packets, subnet 3280 * administration must have created the multicast group and configured 3281 * the fabric appropriately. The port associated with the specified 3282 * QP must also be a member of the multicast group. 3283 */ 3284int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid); 3285 3286/** 3287 * ib_detach_mcast - Detaches the specified QP from a multicast group. 3288 * @qp: QP to detach from the multicast group. 3289 * @gid: Multicast group GID. 3290 * @lid: Multicast group LID in host byte order. 3291 */ 3292int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid); 3293 3294/** 3295 * ib_alloc_xrcd - Allocates an XRC domain. 3296 * @device: The device on which to allocate the XRC domain. 3297 */ 3298struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device); 3299 3300/** 3301 * ib_dealloc_xrcd - Deallocates an XRC domain. 3302 * @xrcd: The XRC domain to deallocate. 3303 */ 3304int ib_dealloc_xrcd(struct ib_xrcd *xrcd); 3305 3306struct ib_flow *ib_create_flow(struct ib_qp *qp, 3307 struct ib_flow_attr *flow_attr, int domain); 3308int ib_destroy_flow(struct ib_flow *flow_id); 3309 3310static inline int ib_check_mr_access(int flags) 3311{ 3312 /* 3313 * Local write permission is required if remote write or 3314 * remote atomic permission is also requested. 3315 */ 3316 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) && 3317 !(flags & IB_ACCESS_LOCAL_WRITE)) 3318 return -EINVAL; 3319 3320 return 0; 3321} 3322 3323/** 3324 * ib_check_mr_status: lightweight check of MR status. 3325 * This routine may provide status checks on a selected 3326 * ib_mr. first use is for signature status check. 3327 * 3328 * @mr: A memory region. 3329 * @check_mask: Bitmask of which checks to perform from 3330 * ib_mr_status_check enumeration. 3331 * @mr_status: The container of relevant status checks. 3332 * failed checks will be indicated in the status bitmask 3333 * and the relevant info shall be in the error item. 3334 */ 3335int ib_check_mr_status(struct ib_mr *mr, u32 check_mask, 3336 struct ib_mr_status *mr_status); 3337 3338struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port, 3339 u16 pkey, const union ib_gid *gid, 3340 const struct sockaddr *addr); 3341struct ib_wq *ib_create_wq(struct ib_pd *pd, 3342 struct ib_wq_init_attr *init_attr); 3343int ib_destroy_wq(struct ib_wq *wq); 3344int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr, 3345 u32 wq_attr_mask); 3346struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device, 3347 struct ib_rwq_ind_table_init_attr* 3348 wq_ind_table_init_attr); 3349int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table); 3350 3351int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents, 3352 unsigned int *sg_offset, unsigned int page_size); 3353 3354static inline int 3355ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents, 3356 unsigned int *sg_offset, unsigned int page_size) 3357{ 3358 int n; 3359 3360 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size); 3361 mr->iova = 0; 3362 3363 return n; 3364} 3365 3366int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents, 3367 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64)); 3368 3369void ib_drain_rq(struct ib_qp *qp); 3370void ib_drain_sq(struct ib_qp *qp); 3371void ib_drain_qp(struct ib_qp *qp); 3372 3373int ib_resolve_eth_dmac(struct ib_device *device, 3374 struct ib_ah_attr *ah_attr); 3375#endif /* IB_VERBS_H */ 3376