1 /* 2 * Copyright (c) 2004 Mellanox Technologies Ltd. All rights reserved. 3 * Copyright (c) 2004 Infinicon Corporation. All rights reserved. 4 * Copyright (c) 2004 Intel Corporation. All rights reserved. 5 * Copyright (c) 2004 Topspin Corporation. All rights reserved. 6 * Copyright (c) 2004 Voltaire Corporation. All rights reserved. 7 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved. 8 * Copyright (c) 2005, 2006, 2007 Cisco Systems. All rights reserved. 9 * 10 * This software is available to you under a choice of one of two 11 * licenses. You may choose to be licensed under the terms of the GNU 12 * General Public License (GPL) Version 2, available from the file 13 * COPYING in the main directory of this source tree, or the 14 * OpenIB.org BSD license below: 15 * 16 * Redistribution and use in source and binary forms, with or 17 * without modification, are permitted provided that the following 18 * conditions are met: 19 * 20 * - Redistributions of source code must retain the above 21 * copyright notice, this list of conditions and the following 22 * disclaimer. 23 * 24 * - Redistributions in binary form must reproduce the above 25 * copyright notice, this list of conditions and the following 26 * disclaimer in the documentation and/or other materials 27 * provided with the distribution. 28 * 29 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 30 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 31 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 32 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 33 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 34 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 35 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 36 * SOFTWARE. 37 */ 38 39 #if !defined(IB_VERBS_H) 40 #define IB_VERBS_H 41 42 #include <linux/types.h> 43 #include <linux/device.h> 44 #include <linux/mm.h> 45 #include <linux/dma-mapping.h> 46 #include <linux/kref.h> 47 #include <linux/list.h> 48 #include <linux/rwsem.h> 49 #include <linux/scatterlist.h> 50 #include <linux/workqueue.h> 51 #include <linux/socket.h> 52 #include <linux/irq_poll.h> 53 #include <uapi/linux/if_ether.h> 54 #include <net/ipv6.h> 55 #include <net/ip.h> 56 #include <linux/string.h> 57 #include <linux/slab.h> 58 #include <linux/netdevice.h> 59 60 #include <linux/if_link.h> 61 #include <linux/atomic.h> 62 #include <linux/mmu_notifier.h> 63 #include <linux/uaccess.h> 64 #include <linux/cgroup_rdma.h> 65 #include <uapi/rdma/ib_user_verbs.h> 66 #include <rdma/restrack.h> 67 #include <uapi/rdma/rdma_user_ioctl.h> 68 #include <uapi/rdma/ib_user_ioctl_verbs.h> 69 70 #define IB_FW_VERSION_NAME_MAX ETHTOOL_FWVERS_LEN 71 72 extern struct workqueue_struct *ib_wq; 73 extern struct workqueue_struct *ib_comp_wq; 74 75 union ib_gid { 76 u8 raw[16]; 77 struct { 78 __be64 subnet_prefix; 79 __be64 interface_id; 80 } global; 81 }; 82 83 extern union ib_gid zgid; 84 85 enum ib_gid_type { 86 /* If link layer is Ethernet, this is RoCE V1 */ 87 IB_GID_TYPE_IB = 0, 88 IB_GID_TYPE_ROCE = 0, 89 IB_GID_TYPE_ROCE_UDP_ENCAP = 1, 90 IB_GID_TYPE_SIZE 91 }; 92 93 #define ROCE_V2_UDP_DPORT 4791 94 struct ib_gid_attr { 95 struct net_device *ndev; 96 struct ib_device *device; 97 union ib_gid gid; 98 enum ib_gid_type gid_type; 99 u16 index; 100 u8 port_num; 101 }; 102 103 enum rdma_node_type { 104 /* IB values map to NodeInfo:NodeType. */ 105 RDMA_NODE_IB_CA = 1, 106 RDMA_NODE_IB_SWITCH, 107 RDMA_NODE_IB_ROUTER, 108 RDMA_NODE_RNIC, 109 RDMA_NODE_USNIC, 110 RDMA_NODE_USNIC_UDP, 111 }; 112 113 enum { 114 /* set the local administered indication */ 115 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2, 116 }; 117 118 enum rdma_transport_type { 119 RDMA_TRANSPORT_IB, 120 RDMA_TRANSPORT_IWARP, 121 RDMA_TRANSPORT_USNIC, 122 RDMA_TRANSPORT_USNIC_UDP 123 }; 124 125 enum rdma_protocol_type { 126 RDMA_PROTOCOL_IB, 127 RDMA_PROTOCOL_IBOE, 128 RDMA_PROTOCOL_IWARP, 129 RDMA_PROTOCOL_USNIC_UDP 130 }; 131 132 __attribute_const__ enum rdma_transport_type 133 rdma_node_get_transport(enum rdma_node_type node_type); 134 135 enum rdma_network_type { 136 RDMA_NETWORK_IB, 137 RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB, 138 RDMA_NETWORK_IPV4, 139 RDMA_NETWORK_IPV6 140 }; 141 142 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type) 143 { 144 if (network_type == RDMA_NETWORK_IPV4 || 145 network_type == RDMA_NETWORK_IPV6) 146 return IB_GID_TYPE_ROCE_UDP_ENCAP; 147 148 /* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */ 149 return IB_GID_TYPE_IB; 150 } 151 152 static inline enum rdma_network_type 153 rdma_gid_attr_network_type(const struct ib_gid_attr *attr) 154 { 155 if (attr->gid_type == IB_GID_TYPE_IB) 156 return RDMA_NETWORK_IB; 157 158 if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid)) 159 return RDMA_NETWORK_IPV4; 160 else 161 return RDMA_NETWORK_IPV6; 162 } 163 164 enum rdma_link_layer { 165 IB_LINK_LAYER_UNSPECIFIED, 166 IB_LINK_LAYER_INFINIBAND, 167 IB_LINK_LAYER_ETHERNET, 168 }; 169 170 enum ib_device_cap_flags { 171 IB_DEVICE_RESIZE_MAX_WR = (1 << 0), 172 IB_DEVICE_BAD_PKEY_CNTR = (1 << 1), 173 IB_DEVICE_BAD_QKEY_CNTR = (1 << 2), 174 IB_DEVICE_RAW_MULTI = (1 << 3), 175 IB_DEVICE_AUTO_PATH_MIG = (1 << 4), 176 IB_DEVICE_CHANGE_PHY_PORT = (1 << 5), 177 IB_DEVICE_UD_AV_PORT_ENFORCE = (1 << 6), 178 IB_DEVICE_CURR_QP_STATE_MOD = (1 << 7), 179 IB_DEVICE_SHUTDOWN_PORT = (1 << 8), 180 /* Not in use, former INIT_TYPE = (1 << 9),*/ 181 IB_DEVICE_PORT_ACTIVE_EVENT = (1 << 10), 182 IB_DEVICE_SYS_IMAGE_GUID = (1 << 11), 183 IB_DEVICE_RC_RNR_NAK_GEN = (1 << 12), 184 IB_DEVICE_SRQ_RESIZE = (1 << 13), 185 IB_DEVICE_N_NOTIFY_CQ = (1 << 14), 186 187 /* 188 * This device supports a per-device lkey or stag that can be 189 * used without performing a memory registration for the local 190 * memory. Note that ULPs should never check this flag, but 191 * instead of use the local_dma_lkey flag in the ib_pd structure, 192 * which will always contain a usable lkey. 193 */ 194 IB_DEVICE_LOCAL_DMA_LKEY = (1 << 15), 195 /* Reserved, old SEND_W_INV = (1 << 16),*/ 196 IB_DEVICE_MEM_WINDOW = (1 << 17), 197 /* 198 * Devices should set IB_DEVICE_UD_IP_SUM if they support 199 * insertion of UDP and TCP checksum on outgoing UD IPoIB 200 * messages and can verify the validity of checksum for 201 * incoming messages. Setting this flag implies that the 202 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode. 203 */ 204 IB_DEVICE_UD_IP_CSUM = (1 << 18), 205 IB_DEVICE_UD_TSO = (1 << 19), 206 IB_DEVICE_XRC = (1 << 20), 207 208 /* 209 * This device supports the IB "base memory management extension", 210 * which includes support for fast registrations (IB_WR_REG_MR, 211 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should 212 * also be set by any iWarp device which must support FRs to comply 213 * to the iWarp verbs spec. iWarp devices also support the 214 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the 215 * stag. 216 */ 217 IB_DEVICE_MEM_MGT_EXTENSIONS = (1 << 21), 218 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1 << 22), 219 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1 << 23), 220 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1 << 24), 221 IB_DEVICE_RC_IP_CSUM = (1 << 25), 222 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */ 223 IB_DEVICE_RAW_IP_CSUM = (1 << 26), 224 /* 225 * Devices should set IB_DEVICE_CROSS_CHANNEL if they 226 * support execution of WQEs that involve synchronization 227 * of I/O operations with single completion queue managed 228 * by hardware. 229 */ 230 IB_DEVICE_CROSS_CHANNEL = (1 << 27), 231 IB_DEVICE_MANAGED_FLOW_STEERING = (1 << 29), 232 IB_DEVICE_SIGNATURE_HANDOVER = (1 << 30), 233 IB_DEVICE_ON_DEMAND_PAGING = (1ULL << 31), 234 IB_DEVICE_SG_GAPS_REG = (1ULL << 32), 235 IB_DEVICE_VIRTUAL_FUNCTION = (1ULL << 33), 236 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */ 237 IB_DEVICE_RAW_SCATTER_FCS = (1ULL << 34), 238 IB_DEVICE_RDMA_NETDEV_OPA_VNIC = (1ULL << 35), 239 /* The device supports padding incoming writes to cacheline. */ 240 IB_DEVICE_PCI_WRITE_END_PADDING = (1ULL << 36), 241 }; 242 243 enum ib_signature_prot_cap { 244 IB_PROT_T10DIF_TYPE_1 = 1, 245 IB_PROT_T10DIF_TYPE_2 = 1 << 1, 246 IB_PROT_T10DIF_TYPE_3 = 1 << 2, 247 }; 248 249 enum ib_signature_guard_cap { 250 IB_GUARD_T10DIF_CRC = 1, 251 IB_GUARD_T10DIF_CSUM = 1 << 1, 252 }; 253 254 enum ib_atomic_cap { 255 IB_ATOMIC_NONE, 256 IB_ATOMIC_HCA, 257 IB_ATOMIC_GLOB 258 }; 259 260 enum ib_odp_general_cap_bits { 261 IB_ODP_SUPPORT = 1 << 0, 262 IB_ODP_SUPPORT_IMPLICIT = 1 << 1, 263 }; 264 265 enum ib_odp_transport_cap_bits { 266 IB_ODP_SUPPORT_SEND = 1 << 0, 267 IB_ODP_SUPPORT_RECV = 1 << 1, 268 IB_ODP_SUPPORT_WRITE = 1 << 2, 269 IB_ODP_SUPPORT_READ = 1 << 3, 270 IB_ODP_SUPPORT_ATOMIC = 1 << 4, 271 }; 272 273 struct ib_odp_caps { 274 uint64_t general_caps; 275 struct { 276 uint32_t rc_odp_caps; 277 uint32_t uc_odp_caps; 278 uint32_t ud_odp_caps; 279 } per_transport_caps; 280 }; 281 282 struct ib_rss_caps { 283 /* Corresponding bit will be set if qp type from 284 * 'enum ib_qp_type' is supported, e.g. 285 * supported_qpts |= 1 << IB_QPT_UD 286 */ 287 u32 supported_qpts; 288 u32 max_rwq_indirection_tables; 289 u32 max_rwq_indirection_table_size; 290 }; 291 292 enum ib_tm_cap_flags { 293 /* Support tag matching on RC transport */ 294 IB_TM_CAP_RC = 1 << 0, 295 }; 296 297 struct ib_tm_caps { 298 /* Max size of RNDV header */ 299 u32 max_rndv_hdr_size; 300 /* Max number of entries in tag matching list */ 301 u32 max_num_tags; 302 /* From enum ib_tm_cap_flags */ 303 u32 flags; 304 /* Max number of outstanding list operations */ 305 u32 max_ops; 306 /* Max number of SGE in tag matching entry */ 307 u32 max_sge; 308 }; 309 310 struct ib_cq_init_attr { 311 unsigned int cqe; 312 int comp_vector; 313 u32 flags; 314 }; 315 316 enum ib_cq_attr_mask { 317 IB_CQ_MODERATE = 1 << 0, 318 }; 319 320 struct ib_cq_caps { 321 u16 max_cq_moderation_count; 322 u16 max_cq_moderation_period; 323 }; 324 325 struct ib_dm_mr_attr { 326 u64 length; 327 u64 offset; 328 u32 access_flags; 329 }; 330 331 struct ib_dm_alloc_attr { 332 u64 length; 333 u32 alignment; 334 u32 flags; 335 }; 336 337 struct ib_device_attr { 338 u64 fw_ver; 339 __be64 sys_image_guid; 340 u64 max_mr_size; 341 u64 page_size_cap; 342 u32 vendor_id; 343 u32 vendor_part_id; 344 u32 hw_ver; 345 int max_qp; 346 int max_qp_wr; 347 u64 device_cap_flags; 348 int max_send_sge; 349 int max_recv_sge; 350 int max_sge_rd; 351 int max_cq; 352 int max_cqe; 353 int max_mr; 354 int max_pd; 355 int max_qp_rd_atom; 356 int max_ee_rd_atom; 357 int max_res_rd_atom; 358 int max_qp_init_rd_atom; 359 int max_ee_init_rd_atom; 360 enum ib_atomic_cap atomic_cap; 361 enum ib_atomic_cap masked_atomic_cap; 362 int max_ee; 363 int max_rdd; 364 int max_mw; 365 int max_raw_ipv6_qp; 366 int max_raw_ethy_qp; 367 int max_mcast_grp; 368 int max_mcast_qp_attach; 369 int max_total_mcast_qp_attach; 370 int max_ah; 371 int max_fmr; 372 int max_map_per_fmr; 373 int max_srq; 374 int max_srq_wr; 375 int max_srq_sge; 376 unsigned int max_fast_reg_page_list_len; 377 u16 max_pkeys; 378 u8 local_ca_ack_delay; 379 int sig_prot_cap; 380 int sig_guard_cap; 381 struct ib_odp_caps odp_caps; 382 uint64_t timestamp_mask; 383 uint64_t hca_core_clock; /* in KHZ */ 384 struct ib_rss_caps rss_caps; 385 u32 max_wq_type_rq; 386 u32 raw_packet_caps; /* Use ib_raw_packet_caps enum */ 387 struct ib_tm_caps tm_caps; 388 struct ib_cq_caps cq_caps; 389 u64 max_dm_size; 390 }; 391 392 enum ib_mtu { 393 IB_MTU_256 = 1, 394 IB_MTU_512 = 2, 395 IB_MTU_1024 = 3, 396 IB_MTU_2048 = 4, 397 IB_MTU_4096 = 5 398 }; 399 400 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu) 401 { 402 switch (mtu) { 403 case IB_MTU_256: return 256; 404 case IB_MTU_512: return 512; 405 case IB_MTU_1024: return 1024; 406 case IB_MTU_2048: return 2048; 407 case IB_MTU_4096: return 4096; 408 default: return -1; 409 } 410 } 411 412 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu) 413 { 414 if (mtu >= 4096) 415 return IB_MTU_4096; 416 else if (mtu >= 2048) 417 return IB_MTU_2048; 418 else if (mtu >= 1024) 419 return IB_MTU_1024; 420 else if (mtu >= 512) 421 return IB_MTU_512; 422 else 423 return IB_MTU_256; 424 } 425 426 enum ib_port_state { 427 IB_PORT_NOP = 0, 428 IB_PORT_DOWN = 1, 429 IB_PORT_INIT = 2, 430 IB_PORT_ARMED = 3, 431 IB_PORT_ACTIVE = 4, 432 IB_PORT_ACTIVE_DEFER = 5 433 }; 434 435 enum ib_port_cap_flags { 436 IB_PORT_SM = 1 << 1, 437 IB_PORT_NOTICE_SUP = 1 << 2, 438 IB_PORT_TRAP_SUP = 1 << 3, 439 IB_PORT_OPT_IPD_SUP = 1 << 4, 440 IB_PORT_AUTO_MIGR_SUP = 1 << 5, 441 IB_PORT_SL_MAP_SUP = 1 << 6, 442 IB_PORT_MKEY_NVRAM = 1 << 7, 443 IB_PORT_PKEY_NVRAM = 1 << 8, 444 IB_PORT_LED_INFO_SUP = 1 << 9, 445 IB_PORT_SM_DISABLED = 1 << 10, 446 IB_PORT_SYS_IMAGE_GUID_SUP = 1 << 11, 447 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP = 1 << 12, 448 IB_PORT_EXTENDED_SPEEDS_SUP = 1 << 14, 449 IB_PORT_CM_SUP = 1 << 16, 450 IB_PORT_SNMP_TUNNEL_SUP = 1 << 17, 451 IB_PORT_REINIT_SUP = 1 << 18, 452 IB_PORT_DEVICE_MGMT_SUP = 1 << 19, 453 IB_PORT_VENDOR_CLASS_SUP = 1 << 20, 454 IB_PORT_DR_NOTICE_SUP = 1 << 21, 455 IB_PORT_CAP_MASK_NOTICE_SUP = 1 << 22, 456 IB_PORT_BOOT_MGMT_SUP = 1 << 23, 457 IB_PORT_LINK_LATENCY_SUP = 1 << 24, 458 IB_PORT_CLIENT_REG_SUP = 1 << 25, 459 IB_PORT_IP_BASED_GIDS = 1 << 26, 460 }; 461 462 enum ib_port_width { 463 IB_WIDTH_1X = 1, 464 IB_WIDTH_4X = 2, 465 IB_WIDTH_8X = 4, 466 IB_WIDTH_12X = 8 467 }; 468 469 static inline int ib_width_enum_to_int(enum ib_port_width width) 470 { 471 switch (width) { 472 case IB_WIDTH_1X: return 1; 473 case IB_WIDTH_4X: return 4; 474 case IB_WIDTH_8X: return 8; 475 case IB_WIDTH_12X: return 12; 476 default: return -1; 477 } 478 } 479 480 enum ib_port_speed { 481 IB_SPEED_SDR = 1, 482 IB_SPEED_DDR = 2, 483 IB_SPEED_QDR = 4, 484 IB_SPEED_FDR10 = 8, 485 IB_SPEED_FDR = 16, 486 IB_SPEED_EDR = 32, 487 IB_SPEED_HDR = 64 488 }; 489 490 /** 491 * struct rdma_hw_stats 492 * @lock - Mutex to protect parallel write access to lifespan and values 493 * of counters, which are 64bits and not guaranteeed to be written 494 * atomicaly on 32bits systems. 495 * @timestamp - Used by the core code to track when the last update was 496 * @lifespan - Used by the core code to determine how old the counters 497 * should be before being updated again. Stored in jiffies, defaults 498 * to 10 milliseconds, drivers can override the default be specifying 499 * their own value during their allocation routine. 500 * @name - Array of pointers to static names used for the counters in 501 * directory. 502 * @num_counters - How many hardware counters there are. If name is 503 * shorter than this number, a kernel oops will result. Driver authors 504 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters) 505 * in their code to prevent this. 506 * @value - Array of u64 counters that are accessed by the sysfs code and 507 * filled in by the drivers get_stats routine 508 */ 509 struct rdma_hw_stats { 510 struct mutex lock; /* Protect lifespan and values[] */ 511 unsigned long timestamp; 512 unsigned long lifespan; 513 const char * const *names; 514 int num_counters; 515 u64 value[]; 516 }; 517 518 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10 519 /** 520 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct 521 * for drivers. 522 * @names - Array of static const char * 523 * @num_counters - How many elements in array 524 * @lifespan - How many milliseconds between updates 525 */ 526 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct( 527 const char * const *names, int num_counters, 528 unsigned long lifespan) 529 { 530 struct rdma_hw_stats *stats; 531 532 stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64), 533 GFP_KERNEL); 534 if (!stats) 535 return NULL; 536 stats->names = names; 537 stats->num_counters = num_counters; 538 stats->lifespan = msecs_to_jiffies(lifespan); 539 540 return stats; 541 } 542 543 544 /* Define bits for the various functionality this port needs to be supported by 545 * the core. 546 */ 547 /* Management 0x00000FFF */ 548 #define RDMA_CORE_CAP_IB_MAD 0x00000001 549 #define RDMA_CORE_CAP_IB_SMI 0x00000002 550 #define RDMA_CORE_CAP_IB_CM 0x00000004 551 #define RDMA_CORE_CAP_IW_CM 0x00000008 552 #define RDMA_CORE_CAP_IB_SA 0x00000010 553 #define RDMA_CORE_CAP_OPA_MAD 0x00000020 554 555 /* Address format 0x000FF000 */ 556 #define RDMA_CORE_CAP_AF_IB 0x00001000 557 #define RDMA_CORE_CAP_ETH_AH 0x00002000 558 #define RDMA_CORE_CAP_OPA_AH 0x00004000 559 560 /* Protocol 0xFFF00000 */ 561 #define RDMA_CORE_CAP_PROT_IB 0x00100000 562 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000 563 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000 564 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000 565 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000 566 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000 567 568 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \ 569 | RDMA_CORE_CAP_IB_MAD \ 570 | RDMA_CORE_CAP_IB_SMI \ 571 | RDMA_CORE_CAP_IB_CM \ 572 | RDMA_CORE_CAP_IB_SA \ 573 | RDMA_CORE_CAP_AF_IB) 574 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \ 575 | RDMA_CORE_CAP_IB_MAD \ 576 | RDMA_CORE_CAP_IB_CM \ 577 | RDMA_CORE_CAP_AF_IB \ 578 | RDMA_CORE_CAP_ETH_AH) 579 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \ 580 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \ 581 | RDMA_CORE_CAP_IB_MAD \ 582 | RDMA_CORE_CAP_IB_CM \ 583 | RDMA_CORE_CAP_AF_IB \ 584 | RDMA_CORE_CAP_ETH_AH) 585 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \ 586 | RDMA_CORE_CAP_IW_CM) 587 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \ 588 | RDMA_CORE_CAP_OPA_MAD) 589 590 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET) 591 592 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC) 593 594 struct ib_port_attr { 595 u64 subnet_prefix; 596 enum ib_port_state state; 597 enum ib_mtu max_mtu; 598 enum ib_mtu active_mtu; 599 int gid_tbl_len; 600 u32 port_cap_flags; 601 u32 max_msg_sz; 602 u32 bad_pkey_cntr; 603 u32 qkey_viol_cntr; 604 u16 pkey_tbl_len; 605 u32 sm_lid; 606 u32 lid; 607 u8 lmc; 608 u8 max_vl_num; 609 u8 sm_sl; 610 u8 subnet_timeout; 611 u8 init_type_reply; 612 u8 active_width; 613 u8 active_speed; 614 u8 phys_state; 615 bool grh_required; 616 }; 617 618 enum ib_device_modify_flags { 619 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0, 620 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1 621 }; 622 623 #define IB_DEVICE_NODE_DESC_MAX 64 624 625 struct ib_device_modify { 626 u64 sys_image_guid; 627 char node_desc[IB_DEVICE_NODE_DESC_MAX]; 628 }; 629 630 enum ib_port_modify_flags { 631 IB_PORT_SHUTDOWN = 1, 632 IB_PORT_INIT_TYPE = (1<<2), 633 IB_PORT_RESET_QKEY_CNTR = (1<<3), 634 IB_PORT_OPA_MASK_CHG = (1<<4) 635 }; 636 637 struct ib_port_modify { 638 u32 set_port_cap_mask; 639 u32 clr_port_cap_mask; 640 u8 init_type; 641 }; 642 643 enum ib_event_type { 644 IB_EVENT_CQ_ERR, 645 IB_EVENT_QP_FATAL, 646 IB_EVENT_QP_REQ_ERR, 647 IB_EVENT_QP_ACCESS_ERR, 648 IB_EVENT_COMM_EST, 649 IB_EVENT_SQ_DRAINED, 650 IB_EVENT_PATH_MIG, 651 IB_EVENT_PATH_MIG_ERR, 652 IB_EVENT_DEVICE_FATAL, 653 IB_EVENT_PORT_ACTIVE, 654 IB_EVENT_PORT_ERR, 655 IB_EVENT_LID_CHANGE, 656 IB_EVENT_PKEY_CHANGE, 657 IB_EVENT_SM_CHANGE, 658 IB_EVENT_SRQ_ERR, 659 IB_EVENT_SRQ_LIMIT_REACHED, 660 IB_EVENT_QP_LAST_WQE_REACHED, 661 IB_EVENT_CLIENT_REREGISTER, 662 IB_EVENT_GID_CHANGE, 663 IB_EVENT_WQ_FATAL, 664 }; 665 666 const char *__attribute_const__ ib_event_msg(enum ib_event_type event); 667 668 struct ib_event { 669 struct ib_device *device; 670 union { 671 struct ib_cq *cq; 672 struct ib_qp *qp; 673 struct ib_srq *srq; 674 struct ib_wq *wq; 675 u8 port_num; 676 } element; 677 enum ib_event_type event; 678 }; 679 680 struct ib_event_handler { 681 struct ib_device *device; 682 void (*handler)(struct ib_event_handler *, struct ib_event *); 683 struct list_head list; 684 }; 685 686 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \ 687 do { \ 688 (_ptr)->device = _device; \ 689 (_ptr)->handler = _handler; \ 690 INIT_LIST_HEAD(&(_ptr)->list); \ 691 } while (0) 692 693 struct ib_global_route { 694 const struct ib_gid_attr *sgid_attr; 695 union ib_gid dgid; 696 u32 flow_label; 697 u8 sgid_index; 698 u8 hop_limit; 699 u8 traffic_class; 700 }; 701 702 struct ib_grh { 703 __be32 version_tclass_flow; 704 __be16 paylen; 705 u8 next_hdr; 706 u8 hop_limit; 707 union ib_gid sgid; 708 union ib_gid dgid; 709 }; 710 711 union rdma_network_hdr { 712 struct ib_grh ibgrh; 713 struct { 714 /* The IB spec states that if it's IPv4, the header 715 * is located in the last 20 bytes of the header. 716 */ 717 u8 reserved[20]; 718 struct iphdr roce4grh; 719 }; 720 }; 721 722 #define IB_QPN_MASK 0xFFFFFF 723 724 enum { 725 IB_MULTICAST_QPN = 0xffffff 726 }; 727 728 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF) 729 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000) 730 731 enum ib_ah_flags { 732 IB_AH_GRH = 1 733 }; 734 735 enum ib_rate { 736 IB_RATE_PORT_CURRENT = 0, 737 IB_RATE_2_5_GBPS = 2, 738 IB_RATE_5_GBPS = 5, 739 IB_RATE_10_GBPS = 3, 740 IB_RATE_20_GBPS = 6, 741 IB_RATE_30_GBPS = 4, 742 IB_RATE_40_GBPS = 7, 743 IB_RATE_60_GBPS = 8, 744 IB_RATE_80_GBPS = 9, 745 IB_RATE_120_GBPS = 10, 746 IB_RATE_14_GBPS = 11, 747 IB_RATE_56_GBPS = 12, 748 IB_RATE_112_GBPS = 13, 749 IB_RATE_168_GBPS = 14, 750 IB_RATE_25_GBPS = 15, 751 IB_RATE_100_GBPS = 16, 752 IB_RATE_200_GBPS = 17, 753 IB_RATE_300_GBPS = 18 754 }; 755 756 /** 757 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the 758 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be 759 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec. 760 * @rate: rate to convert. 761 */ 762 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate); 763 764 /** 765 * ib_rate_to_mbps - Convert the IB rate enum to Mbps. 766 * For example, IB_RATE_2_5_GBPS will be converted to 2500. 767 * @rate: rate to convert. 768 */ 769 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate); 770 771 772 /** 773 * enum ib_mr_type - memory region type 774 * @IB_MR_TYPE_MEM_REG: memory region that is used for 775 * normal registration 776 * @IB_MR_TYPE_SIGNATURE: memory region that is used for 777 * signature operations (data-integrity 778 * capable regions) 779 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to 780 * register any arbitrary sg lists (without 781 * the normal mr constraints - see 782 * ib_map_mr_sg) 783 */ 784 enum ib_mr_type { 785 IB_MR_TYPE_MEM_REG, 786 IB_MR_TYPE_SIGNATURE, 787 IB_MR_TYPE_SG_GAPS, 788 }; 789 790 /** 791 * Signature types 792 * IB_SIG_TYPE_NONE: Unprotected. 793 * IB_SIG_TYPE_T10_DIF: Type T10-DIF 794 */ 795 enum ib_signature_type { 796 IB_SIG_TYPE_NONE, 797 IB_SIG_TYPE_T10_DIF, 798 }; 799 800 /** 801 * Signature T10-DIF block-guard types 802 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules. 803 * IB_T10DIF_CSUM: Corresponds to IP checksum rules. 804 */ 805 enum ib_t10_dif_bg_type { 806 IB_T10DIF_CRC, 807 IB_T10DIF_CSUM 808 }; 809 810 /** 811 * struct ib_t10_dif_domain - Parameters specific for T10-DIF 812 * domain. 813 * @bg_type: T10-DIF block guard type (CRC|CSUM) 814 * @pi_interval: protection information interval. 815 * @bg: seed of guard computation. 816 * @app_tag: application tag of guard block 817 * @ref_tag: initial guard block reference tag. 818 * @ref_remap: Indicate wethear the reftag increments each block 819 * @app_escape: Indicate to skip block check if apptag=0xffff 820 * @ref_escape: Indicate to skip block check if reftag=0xffffffff 821 * @apptag_check_mask: check bitmask of application tag. 822 */ 823 struct ib_t10_dif_domain { 824 enum ib_t10_dif_bg_type bg_type; 825 u16 pi_interval; 826 u16 bg; 827 u16 app_tag; 828 u32 ref_tag; 829 bool ref_remap; 830 bool app_escape; 831 bool ref_escape; 832 u16 apptag_check_mask; 833 }; 834 835 /** 836 * struct ib_sig_domain - Parameters for signature domain 837 * @sig_type: specific signauture type 838 * @sig: union of all signature domain attributes that may 839 * be used to set domain layout. 840 */ 841 struct ib_sig_domain { 842 enum ib_signature_type sig_type; 843 union { 844 struct ib_t10_dif_domain dif; 845 } sig; 846 }; 847 848 /** 849 * struct ib_sig_attrs - Parameters for signature handover operation 850 * @check_mask: bitmask for signature byte check (8 bytes) 851 * @mem: memory domain layout desciptor. 852 * @wire: wire domain layout desciptor. 853 */ 854 struct ib_sig_attrs { 855 u8 check_mask; 856 struct ib_sig_domain mem; 857 struct ib_sig_domain wire; 858 }; 859 860 enum ib_sig_err_type { 861 IB_SIG_BAD_GUARD, 862 IB_SIG_BAD_REFTAG, 863 IB_SIG_BAD_APPTAG, 864 }; 865 866 /** 867 * Signature check masks (8 bytes in total) according to the T10-PI standard: 868 * -------- -------- ------------ 869 * | GUARD | APPTAG | REFTAG | 870 * | 2B | 2B | 4B | 871 * -------- -------- ------------ 872 */ 873 enum { 874 IB_SIG_CHECK_GUARD = 0xc0, 875 IB_SIG_CHECK_APPTAG = 0x30, 876 IB_SIG_CHECK_REFTAG = 0x0f, 877 }; 878 879 /** 880 * struct ib_sig_err - signature error descriptor 881 */ 882 struct ib_sig_err { 883 enum ib_sig_err_type err_type; 884 u32 expected; 885 u32 actual; 886 u64 sig_err_offset; 887 u32 key; 888 }; 889 890 enum ib_mr_status_check { 891 IB_MR_CHECK_SIG_STATUS = 1, 892 }; 893 894 /** 895 * struct ib_mr_status - Memory region status container 896 * 897 * @fail_status: Bitmask of MR checks status. For each 898 * failed check a corresponding status bit is set. 899 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS 900 * failure. 901 */ 902 struct ib_mr_status { 903 u32 fail_status; 904 struct ib_sig_err sig_err; 905 }; 906 907 /** 908 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate 909 * enum. 910 * @mult: multiple to convert. 911 */ 912 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult); 913 914 enum rdma_ah_attr_type { 915 RDMA_AH_ATTR_TYPE_UNDEFINED, 916 RDMA_AH_ATTR_TYPE_IB, 917 RDMA_AH_ATTR_TYPE_ROCE, 918 RDMA_AH_ATTR_TYPE_OPA, 919 }; 920 921 struct ib_ah_attr { 922 u16 dlid; 923 u8 src_path_bits; 924 }; 925 926 struct roce_ah_attr { 927 u8 dmac[ETH_ALEN]; 928 }; 929 930 struct opa_ah_attr { 931 u32 dlid; 932 u8 src_path_bits; 933 bool make_grd; 934 }; 935 936 struct rdma_ah_attr { 937 struct ib_global_route grh; 938 u8 sl; 939 u8 static_rate; 940 u8 port_num; 941 u8 ah_flags; 942 enum rdma_ah_attr_type type; 943 union { 944 struct ib_ah_attr ib; 945 struct roce_ah_attr roce; 946 struct opa_ah_attr opa; 947 }; 948 }; 949 950 enum ib_wc_status { 951 IB_WC_SUCCESS, 952 IB_WC_LOC_LEN_ERR, 953 IB_WC_LOC_QP_OP_ERR, 954 IB_WC_LOC_EEC_OP_ERR, 955 IB_WC_LOC_PROT_ERR, 956 IB_WC_WR_FLUSH_ERR, 957 IB_WC_MW_BIND_ERR, 958 IB_WC_BAD_RESP_ERR, 959 IB_WC_LOC_ACCESS_ERR, 960 IB_WC_REM_INV_REQ_ERR, 961 IB_WC_REM_ACCESS_ERR, 962 IB_WC_REM_OP_ERR, 963 IB_WC_RETRY_EXC_ERR, 964 IB_WC_RNR_RETRY_EXC_ERR, 965 IB_WC_LOC_RDD_VIOL_ERR, 966 IB_WC_REM_INV_RD_REQ_ERR, 967 IB_WC_REM_ABORT_ERR, 968 IB_WC_INV_EECN_ERR, 969 IB_WC_INV_EEC_STATE_ERR, 970 IB_WC_FATAL_ERR, 971 IB_WC_RESP_TIMEOUT_ERR, 972 IB_WC_GENERAL_ERR 973 }; 974 975 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status); 976 977 enum ib_wc_opcode { 978 IB_WC_SEND, 979 IB_WC_RDMA_WRITE, 980 IB_WC_RDMA_READ, 981 IB_WC_COMP_SWAP, 982 IB_WC_FETCH_ADD, 983 IB_WC_LSO, 984 IB_WC_LOCAL_INV, 985 IB_WC_REG_MR, 986 IB_WC_MASKED_COMP_SWAP, 987 IB_WC_MASKED_FETCH_ADD, 988 /* 989 * Set value of IB_WC_RECV so consumers can test if a completion is a 990 * receive by testing (opcode & IB_WC_RECV). 991 */ 992 IB_WC_RECV = 1 << 7, 993 IB_WC_RECV_RDMA_WITH_IMM 994 }; 995 996 enum ib_wc_flags { 997 IB_WC_GRH = 1, 998 IB_WC_WITH_IMM = (1<<1), 999 IB_WC_WITH_INVALIDATE = (1<<2), 1000 IB_WC_IP_CSUM_OK = (1<<3), 1001 IB_WC_WITH_SMAC = (1<<4), 1002 IB_WC_WITH_VLAN = (1<<5), 1003 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6), 1004 }; 1005 1006 struct ib_wc { 1007 union { 1008 u64 wr_id; 1009 struct ib_cqe *wr_cqe; 1010 }; 1011 enum ib_wc_status status; 1012 enum ib_wc_opcode opcode; 1013 u32 vendor_err; 1014 u32 byte_len; 1015 struct ib_qp *qp; 1016 union { 1017 __be32 imm_data; 1018 u32 invalidate_rkey; 1019 } ex; 1020 u32 src_qp; 1021 u32 slid; 1022 int wc_flags; 1023 u16 pkey_index; 1024 u8 sl; 1025 u8 dlid_path_bits; 1026 u8 port_num; /* valid only for DR SMPs on switches */ 1027 u8 smac[ETH_ALEN]; 1028 u16 vlan_id; 1029 u8 network_hdr_type; 1030 }; 1031 1032 enum ib_cq_notify_flags { 1033 IB_CQ_SOLICITED = 1 << 0, 1034 IB_CQ_NEXT_COMP = 1 << 1, 1035 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP, 1036 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2, 1037 }; 1038 1039 enum ib_srq_type { 1040 IB_SRQT_BASIC, 1041 IB_SRQT_XRC, 1042 IB_SRQT_TM, 1043 }; 1044 1045 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type) 1046 { 1047 return srq_type == IB_SRQT_XRC || 1048 srq_type == IB_SRQT_TM; 1049 } 1050 1051 enum ib_srq_attr_mask { 1052 IB_SRQ_MAX_WR = 1 << 0, 1053 IB_SRQ_LIMIT = 1 << 1, 1054 }; 1055 1056 struct ib_srq_attr { 1057 u32 max_wr; 1058 u32 max_sge; 1059 u32 srq_limit; 1060 }; 1061 1062 struct ib_srq_init_attr { 1063 void (*event_handler)(struct ib_event *, void *); 1064 void *srq_context; 1065 struct ib_srq_attr attr; 1066 enum ib_srq_type srq_type; 1067 1068 struct { 1069 struct ib_cq *cq; 1070 union { 1071 struct { 1072 struct ib_xrcd *xrcd; 1073 } xrc; 1074 1075 struct { 1076 u32 max_num_tags; 1077 } tag_matching; 1078 }; 1079 } ext; 1080 }; 1081 1082 struct ib_qp_cap { 1083 u32 max_send_wr; 1084 u32 max_recv_wr; 1085 u32 max_send_sge; 1086 u32 max_recv_sge; 1087 u32 max_inline_data; 1088 1089 /* 1090 * Maximum number of rdma_rw_ctx structures in flight at a time. 1091 * ib_create_qp() will calculate the right amount of neededed WRs 1092 * and MRs based on this. 1093 */ 1094 u32 max_rdma_ctxs; 1095 }; 1096 1097 enum ib_sig_type { 1098 IB_SIGNAL_ALL_WR, 1099 IB_SIGNAL_REQ_WR 1100 }; 1101 1102 enum ib_qp_type { 1103 /* 1104 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries 1105 * here (and in that order) since the MAD layer uses them as 1106 * indices into a 2-entry table. 1107 */ 1108 IB_QPT_SMI, 1109 IB_QPT_GSI, 1110 1111 IB_QPT_RC, 1112 IB_QPT_UC, 1113 IB_QPT_UD, 1114 IB_QPT_RAW_IPV6, 1115 IB_QPT_RAW_ETHERTYPE, 1116 IB_QPT_RAW_PACKET = 8, 1117 IB_QPT_XRC_INI = 9, 1118 IB_QPT_XRC_TGT, 1119 IB_QPT_MAX, 1120 IB_QPT_DRIVER = 0xFF, 1121 /* Reserve a range for qp types internal to the low level driver. 1122 * These qp types will not be visible at the IB core layer, so the 1123 * IB_QPT_MAX usages should not be affected in the core layer 1124 */ 1125 IB_QPT_RESERVED1 = 0x1000, 1126 IB_QPT_RESERVED2, 1127 IB_QPT_RESERVED3, 1128 IB_QPT_RESERVED4, 1129 IB_QPT_RESERVED5, 1130 IB_QPT_RESERVED6, 1131 IB_QPT_RESERVED7, 1132 IB_QPT_RESERVED8, 1133 IB_QPT_RESERVED9, 1134 IB_QPT_RESERVED10, 1135 }; 1136 1137 enum ib_qp_create_flags { 1138 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0, 1139 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1, 1140 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2, 1141 IB_QP_CREATE_MANAGED_SEND = 1 << 3, 1142 IB_QP_CREATE_MANAGED_RECV = 1 << 4, 1143 IB_QP_CREATE_NETIF_QP = 1 << 5, 1144 IB_QP_CREATE_SIGNATURE_EN = 1 << 6, 1145 /* FREE = 1 << 7, */ 1146 IB_QP_CREATE_SCATTER_FCS = 1 << 8, 1147 IB_QP_CREATE_CVLAN_STRIPPING = 1 << 9, 1148 IB_QP_CREATE_SOURCE_QPN = 1 << 10, 1149 IB_QP_CREATE_PCI_WRITE_END_PADDING = 1 << 11, 1150 /* reserve bits 26-31 for low level drivers' internal use */ 1151 IB_QP_CREATE_RESERVED_START = 1 << 26, 1152 IB_QP_CREATE_RESERVED_END = 1 << 31, 1153 }; 1154 1155 /* 1156 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler 1157 * callback to destroy the passed in QP. 1158 */ 1159 1160 struct ib_qp_init_attr { 1161 void (*event_handler)(struct ib_event *, void *); 1162 void *qp_context; 1163 struct ib_cq *send_cq; 1164 struct ib_cq *recv_cq; 1165 struct ib_srq *srq; 1166 struct ib_xrcd *xrcd; /* XRC TGT QPs only */ 1167 struct ib_qp_cap cap; 1168 enum ib_sig_type sq_sig_type; 1169 enum ib_qp_type qp_type; 1170 enum ib_qp_create_flags create_flags; 1171 1172 /* 1173 * Only needed for special QP types, or when using the RW API. 1174 */ 1175 u8 port_num; 1176 struct ib_rwq_ind_table *rwq_ind_tbl; 1177 u32 source_qpn; 1178 }; 1179 1180 struct ib_qp_open_attr { 1181 void (*event_handler)(struct ib_event *, void *); 1182 void *qp_context; 1183 u32 qp_num; 1184 enum ib_qp_type qp_type; 1185 }; 1186 1187 enum ib_rnr_timeout { 1188 IB_RNR_TIMER_655_36 = 0, 1189 IB_RNR_TIMER_000_01 = 1, 1190 IB_RNR_TIMER_000_02 = 2, 1191 IB_RNR_TIMER_000_03 = 3, 1192 IB_RNR_TIMER_000_04 = 4, 1193 IB_RNR_TIMER_000_06 = 5, 1194 IB_RNR_TIMER_000_08 = 6, 1195 IB_RNR_TIMER_000_12 = 7, 1196 IB_RNR_TIMER_000_16 = 8, 1197 IB_RNR_TIMER_000_24 = 9, 1198 IB_RNR_TIMER_000_32 = 10, 1199 IB_RNR_TIMER_000_48 = 11, 1200 IB_RNR_TIMER_000_64 = 12, 1201 IB_RNR_TIMER_000_96 = 13, 1202 IB_RNR_TIMER_001_28 = 14, 1203 IB_RNR_TIMER_001_92 = 15, 1204 IB_RNR_TIMER_002_56 = 16, 1205 IB_RNR_TIMER_003_84 = 17, 1206 IB_RNR_TIMER_005_12 = 18, 1207 IB_RNR_TIMER_007_68 = 19, 1208 IB_RNR_TIMER_010_24 = 20, 1209 IB_RNR_TIMER_015_36 = 21, 1210 IB_RNR_TIMER_020_48 = 22, 1211 IB_RNR_TIMER_030_72 = 23, 1212 IB_RNR_TIMER_040_96 = 24, 1213 IB_RNR_TIMER_061_44 = 25, 1214 IB_RNR_TIMER_081_92 = 26, 1215 IB_RNR_TIMER_122_88 = 27, 1216 IB_RNR_TIMER_163_84 = 28, 1217 IB_RNR_TIMER_245_76 = 29, 1218 IB_RNR_TIMER_327_68 = 30, 1219 IB_RNR_TIMER_491_52 = 31 1220 }; 1221 1222 enum ib_qp_attr_mask { 1223 IB_QP_STATE = 1, 1224 IB_QP_CUR_STATE = (1<<1), 1225 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2), 1226 IB_QP_ACCESS_FLAGS = (1<<3), 1227 IB_QP_PKEY_INDEX = (1<<4), 1228 IB_QP_PORT = (1<<5), 1229 IB_QP_QKEY = (1<<6), 1230 IB_QP_AV = (1<<7), 1231 IB_QP_PATH_MTU = (1<<8), 1232 IB_QP_TIMEOUT = (1<<9), 1233 IB_QP_RETRY_CNT = (1<<10), 1234 IB_QP_RNR_RETRY = (1<<11), 1235 IB_QP_RQ_PSN = (1<<12), 1236 IB_QP_MAX_QP_RD_ATOMIC = (1<<13), 1237 IB_QP_ALT_PATH = (1<<14), 1238 IB_QP_MIN_RNR_TIMER = (1<<15), 1239 IB_QP_SQ_PSN = (1<<16), 1240 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17), 1241 IB_QP_PATH_MIG_STATE = (1<<18), 1242 IB_QP_CAP = (1<<19), 1243 IB_QP_DEST_QPN = (1<<20), 1244 IB_QP_RESERVED1 = (1<<21), 1245 IB_QP_RESERVED2 = (1<<22), 1246 IB_QP_RESERVED3 = (1<<23), 1247 IB_QP_RESERVED4 = (1<<24), 1248 IB_QP_RATE_LIMIT = (1<<25), 1249 }; 1250 1251 enum ib_qp_state { 1252 IB_QPS_RESET, 1253 IB_QPS_INIT, 1254 IB_QPS_RTR, 1255 IB_QPS_RTS, 1256 IB_QPS_SQD, 1257 IB_QPS_SQE, 1258 IB_QPS_ERR 1259 }; 1260 1261 enum ib_mig_state { 1262 IB_MIG_MIGRATED, 1263 IB_MIG_REARM, 1264 IB_MIG_ARMED 1265 }; 1266 1267 enum ib_mw_type { 1268 IB_MW_TYPE_1 = 1, 1269 IB_MW_TYPE_2 = 2 1270 }; 1271 1272 struct ib_qp_attr { 1273 enum ib_qp_state qp_state; 1274 enum ib_qp_state cur_qp_state; 1275 enum ib_mtu path_mtu; 1276 enum ib_mig_state path_mig_state; 1277 u32 qkey; 1278 u32 rq_psn; 1279 u32 sq_psn; 1280 u32 dest_qp_num; 1281 int qp_access_flags; 1282 struct ib_qp_cap cap; 1283 struct rdma_ah_attr ah_attr; 1284 struct rdma_ah_attr alt_ah_attr; 1285 u16 pkey_index; 1286 u16 alt_pkey_index; 1287 u8 en_sqd_async_notify; 1288 u8 sq_draining; 1289 u8 max_rd_atomic; 1290 u8 max_dest_rd_atomic; 1291 u8 min_rnr_timer; 1292 u8 port_num; 1293 u8 timeout; 1294 u8 retry_cnt; 1295 u8 rnr_retry; 1296 u8 alt_port_num; 1297 u8 alt_timeout; 1298 u32 rate_limit; 1299 }; 1300 1301 enum ib_wr_opcode { 1302 IB_WR_RDMA_WRITE, 1303 IB_WR_RDMA_WRITE_WITH_IMM, 1304 IB_WR_SEND, 1305 IB_WR_SEND_WITH_IMM, 1306 IB_WR_RDMA_READ, 1307 IB_WR_ATOMIC_CMP_AND_SWP, 1308 IB_WR_ATOMIC_FETCH_AND_ADD, 1309 IB_WR_LSO, 1310 IB_WR_SEND_WITH_INV, 1311 IB_WR_RDMA_READ_WITH_INV, 1312 IB_WR_LOCAL_INV, 1313 IB_WR_REG_MR, 1314 IB_WR_MASKED_ATOMIC_CMP_AND_SWP, 1315 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD, 1316 IB_WR_REG_SIG_MR, 1317 /* reserve values for low level drivers' internal use. 1318 * These values will not be used at all in the ib core layer. 1319 */ 1320 IB_WR_RESERVED1 = 0xf0, 1321 IB_WR_RESERVED2, 1322 IB_WR_RESERVED3, 1323 IB_WR_RESERVED4, 1324 IB_WR_RESERVED5, 1325 IB_WR_RESERVED6, 1326 IB_WR_RESERVED7, 1327 IB_WR_RESERVED8, 1328 IB_WR_RESERVED9, 1329 IB_WR_RESERVED10, 1330 }; 1331 1332 enum ib_send_flags { 1333 IB_SEND_FENCE = 1, 1334 IB_SEND_SIGNALED = (1<<1), 1335 IB_SEND_SOLICITED = (1<<2), 1336 IB_SEND_INLINE = (1<<3), 1337 IB_SEND_IP_CSUM = (1<<4), 1338 1339 /* reserve bits 26-31 for low level drivers' internal use */ 1340 IB_SEND_RESERVED_START = (1 << 26), 1341 IB_SEND_RESERVED_END = (1 << 31), 1342 }; 1343 1344 struct ib_sge { 1345 u64 addr; 1346 u32 length; 1347 u32 lkey; 1348 }; 1349 1350 struct ib_cqe { 1351 void (*done)(struct ib_cq *cq, struct ib_wc *wc); 1352 }; 1353 1354 struct ib_send_wr { 1355 struct ib_send_wr *next; 1356 union { 1357 u64 wr_id; 1358 struct ib_cqe *wr_cqe; 1359 }; 1360 struct ib_sge *sg_list; 1361 int num_sge; 1362 enum ib_wr_opcode opcode; 1363 int send_flags; 1364 union { 1365 __be32 imm_data; 1366 u32 invalidate_rkey; 1367 } ex; 1368 }; 1369 1370 struct ib_rdma_wr { 1371 struct ib_send_wr wr; 1372 u64 remote_addr; 1373 u32 rkey; 1374 }; 1375 1376 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr) 1377 { 1378 return container_of(wr, struct ib_rdma_wr, wr); 1379 } 1380 1381 struct ib_atomic_wr { 1382 struct ib_send_wr wr; 1383 u64 remote_addr; 1384 u64 compare_add; 1385 u64 swap; 1386 u64 compare_add_mask; 1387 u64 swap_mask; 1388 u32 rkey; 1389 }; 1390 1391 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr) 1392 { 1393 return container_of(wr, struct ib_atomic_wr, wr); 1394 } 1395 1396 struct ib_ud_wr { 1397 struct ib_send_wr wr; 1398 struct ib_ah *ah; 1399 void *header; 1400 int hlen; 1401 int mss; 1402 u32 remote_qpn; 1403 u32 remote_qkey; 1404 u16 pkey_index; /* valid for GSI only */ 1405 u8 port_num; /* valid for DR SMPs on switch only */ 1406 }; 1407 1408 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr) 1409 { 1410 return container_of(wr, struct ib_ud_wr, wr); 1411 } 1412 1413 struct ib_reg_wr { 1414 struct ib_send_wr wr; 1415 struct ib_mr *mr; 1416 u32 key; 1417 int access; 1418 }; 1419 1420 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr) 1421 { 1422 return container_of(wr, struct ib_reg_wr, wr); 1423 } 1424 1425 struct ib_sig_handover_wr { 1426 struct ib_send_wr wr; 1427 struct ib_sig_attrs *sig_attrs; 1428 struct ib_mr *sig_mr; 1429 int access_flags; 1430 struct ib_sge *prot; 1431 }; 1432 1433 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr) 1434 { 1435 return container_of(wr, struct ib_sig_handover_wr, wr); 1436 } 1437 1438 struct ib_recv_wr { 1439 struct ib_recv_wr *next; 1440 union { 1441 u64 wr_id; 1442 struct ib_cqe *wr_cqe; 1443 }; 1444 struct ib_sge *sg_list; 1445 int num_sge; 1446 }; 1447 1448 enum ib_access_flags { 1449 IB_ACCESS_LOCAL_WRITE = 1, 1450 IB_ACCESS_REMOTE_WRITE = (1<<1), 1451 IB_ACCESS_REMOTE_READ = (1<<2), 1452 IB_ACCESS_REMOTE_ATOMIC = (1<<3), 1453 IB_ACCESS_MW_BIND = (1<<4), 1454 IB_ZERO_BASED = (1<<5), 1455 IB_ACCESS_ON_DEMAND = (1<<6), 1456 IB_ACCESS_HUGETLB = (1<<7), 1457 }; 1458 1459 /* 1460 * XXX: these are apparently used for ->rereg_user_mr, no idea why they 1461 * are hidden here instead of a uapi header! 1462 */ 1463 enum ib_mr_rereg_flags { 1464 IB_MR_REREG_TRANS = 1, 1465 IB_MR_REREG_PD = (1<<1), 1466 IB_MR_REREG_ACCESS = (1<<2), 1467 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1) 1468 }; 1469 1470 struct ib_fmr_attr { 1471 int max_pages; 1472 int max_maps; 1473 u8 page_shift; 1474 }; 1475 1476 struct ib_umem; 1477 1478 enum rdma_remove_reason { 1479 /* Userspace requested uobject deletion. Call could fail */ 1480 RDMA_REMOVE_DESTROY, 1481 /* Context deletion. This call should delete the actual object itself */ 1482 RDMA_REMOVE_CLOSE, 1483 /* Driver is being hot-unplugged. This call should delete the actual object itself */ 1484 RDMA_REMOVE_DRIVER_REMOVE, 1485 /* Context is being cleaned-up, but commit was just completed */ 1486 RDMA_REMOVE_DURING_CLEANUP, 1487 }; 1488 1489 struct ib_rdmacg_object { 1490 #ifdef CONFIG_CGROUP_RDMA 1491 struct rdma_cgroup *cg; /* owner rdma cgroup */ 1492 #endif 1493 }; 1494 1495 struct ib_ucontext { 1496 struct ib_device *device; 1497 struct ib_uverbs_file *ufile; 1498 int closing; 1499 1500 /* locking the uobjects_list */ 1501 struct mutex uobjects_lock; 1502 struct list_head uobjects; 1503 /* protects cleanup process from other actions */ 1504 struct rw_semaphore cleanup_rwsem; 1505 enum rdma_remove_reason cleanup_reason; 1506 1507 struct pid *tgid; 1508 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING 1509 struct rb_root_cached umem_tree; 1510 /* 1511 * Protects .umem_rbroot and tree, as well as odp_mrs_count and 1512 * mmu notifiers registration. 1513 */ 1514 struct rw_semaphore umem_rwsem; 1515 void (*invalidate_range)(struct ib_umem *umem, 1516 unsigned long start, unsigned long end); 1517 1518 struct mmu_notifier mn; 1519 atomic_t notifier_count; 1520 /* A list of umems that don't have private mmu notifier counters yet. */ 1521 struct list_head no_private_counters; 1522 int odp_mrs_count; 1523 #endif 1524 1525 struct ib_rdmacg_object cg_obj; 1526 }; 1527 1528 struct ib_uobject { 1529 u64 user_handle; /* handle given to us by userspace */ 1530 struct ib_ucontext *context; /* associated user context */ 1531 void *object; /* containing object */ 1532 struct list_head list; /* link to context's list */ 1533 struct ib_rdmacg_object cg_obj; /* rdmacg object */ 1534 int id; /* index into kernel idr */ 1535 struct kref ref; 1536 atomic_t usecnt; /* protects exclusive access */ 1537 struct rcu_head rcu; /* kfree_rcu() overhead */ 1538 1539 const struct uverbs_obj_type *type; 1540 }; 1541 1542 struct ib_uobject_file { 1543 struct ib_uobject uobj; 1544 /* ufile contains the lock between context release and file close */ 1545 struct ib_uverbs_file *ufile; 1546 }; 1547 1548 struct ib_udata { 1549 const void __user *inbuf; 1550 void __user *outbuf; 1551 size_t inlen; 1552 size_t outlen; 1553 }; 1554 1555 struct ib_pd { 1556 u32 local_dma_lkey; 1557 u32 flags; 1558 struct ib_device *device; 1559 struct ib_uobject *uobject; 1560 atomic_t usecnt; /* count all resources */ 1561 1562 u32 unsafe_global_rkey; 1563 1564 /* 1565 * Implementation details of the RDMA core, don't use in drivers: 1566 */ 1567 struct ib_mr *__internal_mr; 1568 struct rdma_restrack_entry res; 1569 }; 1570 1571 struct ib_xrcd { 1572 struct ib_device *device; 1573 atomic_t usecnt; /* count all exposed resources */ 1574 struct inode *inode; 1575 1576 struct mutex tgt_qp_mutex; 1577 struct list_head tgt_qp_list; 1578 }; 1579 1580 struct ib_ah { 1581 struct ib_device *device; 1582 struct ib_pd *pd; 1583 struct ib_uobject *uobject; 1584 const struct ib_gid_attr *sgid_attr; 1585 enum rdma_ah_attr_type type; 1586 }; 1587 1588 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context); 1589 1590 enum ib_poll_context { 1591 IB_POLL_DIRECT, /* caller context, no hw completions */ 1592 IB_POLL_SOFTIRQ, /* poll from softirq context */ 1593 IB_POLL_WORKQUEUE, /* poll from workqueue */ 1594 }; 1595 1596 struct ib_cq { 1597 struct ib_device *device; 1598 struct ib_uobject *uobject; 1599 ib_comp_handler comp_handler; 1600 void (*event_handler)(struct ib_event *, void *); 1601 void *cq_context; 1602 int cqe; 1603 atomic_t usecnt; /* count number of work queues */ 1604 enum ib_poll_context poll_ctx; 1605 struct ib_wc *wc; 1606 union { 1607 struct irq_poll iop; 1608 struct work_struct work; 1609 }; 1610 /* 1611 * Implementation details of the RDMA core, don't use in drivers: 1612 */ 1613 struct rdma_restrack_entry res; 1614 }; 1615 1616 struct ib_srq { 1617 struct ib_device *device; 1618 struct ib_pd *pd; 1619 struct ib_uobject *uobject; 1620 void (*event_handler)(struct ib_event *, void *); 1621 void *srq_context; 1622 enum ib_srq_type srq_type; 1623 atomic_t usecnt; 1624 1625 struct { 1626 struct ib_cq *cq; 1627 union { 1628 struct { 1629 struct ib_xrcd *xrcd; 1630 u32 srq_num; 1631 } xrc; 1632 }; 1633 } ext; 1634 }; 1635 1636 enum ib_raw_packet_caps { 1637 /* Strip cvlan from incoming packet and report it in the matching work 1638 * completion is supported. 1639 */ 1640 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = (1 << 0), 1641 /* Scatter FCS field of an incoming packet to host memory is supported. 1642 */ 1643 IB_RAW_PACKET_CAP_SCATTER_FCS = (1 << 1), 1644 /* Checksum offloads are supported (for both send and receive). */ 1645 IB_RAW_PACKET_CAP_IP_CSUM = (1 << 2), 1646 /* When a packet is received for an RQ with no receive WQEs, the 1647 * packet processing is delayed. 1648 */ 1649 IB_RAW_PACKET_CAP_DELAY_DROP = (1 << 3), 1650 }; 1651 1652 enum ib_wq_type { 1653 IB_WQT_RQ 1654 }; 1655 1656 enum ib_wq_state { 1657 IB_WQS_RESET, 1658 IB_WQS_RDY, 1659 IB_WQS_ERR 1660 }; 1661 1662 struct ib_wq { 1663 struct ib_device *device; 1664 struct ib_uobject *uobject; 1665 void *wq_context; 1666 void (*event_handler)(struct ib_event *, void *); 1667 struct ib_pd *pd; 1668 struct ib_cq *cq; 1669 u32 wq_num; 1670 enum ib_wq_state state; 1671 enum ib_wq_type wq_type; 1672 atomic_t usecnt; 1673 }; 1674 1675 enum ib_wq_flags { 1676 IB_WQ_FLAGS_CVLAN_STRIPPING = 1 << 0, 1677 IB_WQ_FLAGS_SCATTER_FCS = 1 << 1, 1678 IB_WQ_FLAGS_DELAY_DROP = 1 << 2, 1679 IB_WQ_FLAGS_PCI_WRITE_END_PADDING = 1 << 3, 1680 }; 1681 1682 struct ib_wq_init_attr { 1683 void *wq_context; 1684 enum ib_wq_type wq_type; 1685 u32 max_wr; 1686 u32 max_sge; 1687 struct ib_cq *cq; 1688 void (*event_handler)(struct ib_event *, void *); 1689 u32 create_flags; /* Use enum ib_wq_flags */ 1690 }; 1691 1692 enum ib_wq_attr_mask { 1693 IB_WQ_STATE = 1 << 0, 1694 IB_WQ_CUR_STATE = 1 << 1, 1695 IB_WQ_FLAGS = 1 << 2, 1696 }; 1697 1698 struct ib_wq_attr { 1699 enum ib_wq_state wq_state; 1700 enum ib_wq_state curr_wq_state; 1701 u32 flags; /* Use enum ib_wq_flags */ 1702 u32 flags_mask; /* Use enum ib_wq_flags */ 1703 }; 1704 1705 struct ib_rwq_ind_table { 1706 struct ib_device *device; 1707 struct ib_uobject *uobject; 1708 atomic_t usecnt; 1709 u32 ind_tbl_num; 1710 u32 log_ind_tbl_size; 1711 struct ib_wq **ind_tbl; 1712 }; 1713 1714 struct ib_rwq_ind_table_init_attr { 1715 u32 log_ind_tbl_size; 1716 /* Each entry is a pointer to Receive Work Queue */ 1717 struct ib_wq **ind_tbl; 1718 }; 1719 1720 enum port_pkey_state { 1721 IB_PORT_PKEY_NOT_VALID = 0, 1722 IB_PORT_PKEY_VALID = 1, 1723 IB_PORT_PKEY_LISTED = 2, 1724 }; 1725 1726 struct ib_qp_security; 1727 1728 struct ib_port_pkey { 1729 enum port_pkey_state state; 1730 u16 pkey_index; 1731 u8 port_num; 1732 struct list_head qp_list; 1733 struct list_head to_error_list; 1734 struct ib_qp_security *sec; 1735 }; 1736 1737 struct ib_ports_pkeys { 1738 struct ib_port_pkey main; 1739 struct ib_port_pkey alt; 1740 }; 1741 1742 struct ib_qp_security { 1743 struct ib_qp *qp; 1744 struct ib_device *dev; 1745 /* Hold this mutex when changing port and pkey settings. */ 1746 struct mutex mutex; 1747 struct ib_ports_pkeys *ports_pkeys; 1748 /* A list of all open shared QP handles. Required to enforce security 1749 * properly for all users of a shared QP. 1750 */ 1751 struct list_head shared_qp_list; 1752 void *security; 1753 bool destroying; 1754 atomic_t error_list_count; 1755 struct completion error_complete; 1756 int error_comps_pending; 1757 }; 1758 1759 /* 1760 * @max_write_sge: Maximum SGE elements per RDMA WRITE request. 1761 * @max_read_sge: Maximum SGE elements per RDMA READ request. 1762 */ 1763 struct ib_qp { 1764 struct ib_device *device; 1765 struct ib_pd *pd; 1766 struct ib_cq *send_cq; 1767 struct ib_cq *recv_cq; 1768 spinlock_t mr_lock; 1769 int mrs_used; 1770 struct list_head rdma_mrs; 1771 struct list_head sig_mrs; 1772 struct ib_srq *srq; 1773 struct ib_xrcd *xrcd; /* XRC TGT QPs only */ 1774 struct list_head xrcd_list; 1775 1776 /* count times opened, mcast attaches, flow attaches */ 1777 atomic_t usecnt; 1778 struct list_head open_list; 1779 struct ib_qp *real_qp; 1780 struct ib_uobject *uobject; 1781 void (*event_handler)(struct ib_event *, void *); 1782 void *qp_context; 1783 /* sgid_attrs associated with the AV's */ 1784 const struct ib_gid_attr *av_sgid_attr; 1785 const struct ib_gid_attr *alt_path_sgid_attr; 1786 u32 qp_num; 1787 u32 max_write_sge; 1788 u32 max_read_sge; 1789 enum ib_qp_type qp_type; 1790 struct ib_rwq_ind_table *rwq_ind_tbl; 1791 struct ib_qp_security *qp_sec; 1792 u8 port; 1793 1794 /* 1795 * Implementation details of the RDMA core, don't use in drivers: 1796 */ 1797 struct rdma_restrack_entry res; 1798 }; 1799 1800 struct ib_dm { 1801 struct ib_device *device; 1802 u32 length; 1803 u32 flags; 1804 struct ib_uobject *uobject; 1805 atomic_t usecnt; 1806 }; 1807 1808 struct ib_mr { 1809 struct ib_device *device; 1810 struct ib_pd *pd; 1811 u32 lkey; 1812 u32 rkey; 1813 u64 iova; 1814 u64 length; 1815 unsigned int page_size; 1816 bool need_inval; 1817 union { 1818 struct ib_uobject *uobject; /* user */ 1819 struct list_head qp_entry; /* FR */ 1820 }; 1821 1822 struct ib_dm *dm; 1823 1824 /* 1825 * Implementation details of the RDMA core, don't use in drivers: 1826 */ 1827 struct rdma_restrack_entry res; 1828 }; 1829 1830 struct ib_mw { 1831 struct ib_device *device; 1832 struct ib_pd *pd; 1833 struct ib_uobject *uobject; 1834 u32 rkey; 1835 enum ib_mw_type type; 1836 }; 1837 1838 struct ib_fmr { 1839 struct ib_device *device; 1840 struct ib_pd *pd; 1841 struct list_head list; 1842 u32 lkey; 1843 u32 rkey; 1844 }; 1845 1846 /* Supported steering options */ 1847 enum ib_flow_attr_type { 1848 /* steering according to rule specifications */ 1849 IB_FLOW_ATTR_NORMAL = 0x0, 1850 /* default unicast and multicast rule - 1851 * receive all Eth traffic which isn't steered to any QP 1852 */ 1853 IB_FLOW_ATTR_ALL_DEFAULT = 0x1, 1854 /* default multicast rule - 1855 * receive all Eth multicast traffic which isn't steered to any QP 1856 */ 1857 IB_FLOW_ATTR_MC_DEFAULT = 0x2, 1858 /* sniffer rule - receive all port traffic */ 1859 IB_FLOW_ATTR_SNIFFER = 0x3 1860 }; 1861 1862 /* Supported steering header types */ 1863 enum ib_flow_spec_type { 1864 /* L2 headers*/ 1865 IB_FLOW_SPEC_ETH = 0x20, 1866 IB_FLOW_SPEC_IB = 0x22, 1867 /* L3 header*/ 1868 IB_FLOW_SPEC_IPV4 = 0x30, 1869 IB_FLOW_SPEC_IPV6 = 0x31, 1870 IB_FLOW_SPEC_ESP = 0x34, 1871 /* L4 headers*/ 1872 IB_FLOW_SPEC_TCP = 0x40, 1873 IB_FLOW_SPEC_UDP = 0x41, 1874 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50, 1875 IB_FLOW_SPEC_GRE = 0x51, 1876 IB_FLOW_SPEC_MPLS = 0x60, 1877 IB_FLOW_SPEC_INNER = 0x100, 1878 /* Actions */ 1879 IB_FLOW_SPEC_ACTION_TAG = 0x1000, 1880 IB_FLOW_SPEC_ACTION_DROP = 0x1001, 1881 IB_FLOW_SPEC_ACTION_HANDLE = 0x1002, 1882 IB_FLOW_SPEC_ACTION_COUNT = 0x1003, 1883 }; 1884 #define IB_FLOW_SPEC_LAYER_MASK 0xF0 1885 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10 1886 1887 /* Flow steering rule priority is set according to it's domain. 1888 * Lower domain value means higher priority. 1889 */ 1890 enum ib_flow_domain { 1891 IB_FLOW_DOMAIN_USER, 1892 IB_FLOW_DOMAIN_ETHTOOL, 1893 IB_FLOW_DOMAIN_RFS, 1894 IB_FLOW_DOMAIN_NIC, 1895 IB_FLOW_DOMAIN_NUM /* Must be last */ 1896 }; 1897 1898 enum ib_flow_flags { 1899 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */ 1900 IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */ 1901 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 3 /* Must be last */ 1902 }; 1903 1904 struct ib_flow_eth_filter { 1905 u8 dst_mac[6]; 1906 u8 src_mac[6]; 1907 __be16 ether_type; 1908 __be16 vlan_tag; 1909 /* Must be last */ 1910 u8 real_sz[0]; 1911 }; 1912 1913 struct ib_flow_spec_eth { 1914 u32 type; 1915 u16 size; 1916 struct ib_flow_eth_filter val; 1917 struct ib_flow_eth_filter mask; 1918 }; 1919 1920 struct ib_flow_ib_filter { 1921 __be16 dlid; 1922 __u8 sl; 1923 /* Must be last */ 1924 u8 real_sz[0]; 1925 }; 1926 1927 struct ib_flow_spec_ib { 1928 u32 type; 1929 u16 size; 1930 struct ib_flow_ib_filter val; 1931 struct ib_flow_ib_filter mask; 1932 }; 1933 1934 /* IPv4 header flags */ 1935 enum ib_ipv4_flags { 1936 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */ 1937 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the 1938 last have this flag set */ 1939 }; 1940 1941 struct ib_flow_ipv4_filter { 1942 __be32 src_ip; 1943 __be32 dst_ip; 1944 u8 proto; 1945 u8 tos; 1946 u8 ttl; 1947 u8 flags; 1948 /* Must be last */ 1949 u8 real_sz[0]; 1950 }; 1951 1952 struct ib_flow_spec_ipv4 { 1953 u32 type; 1954 u16 size; 1955 struct ib_flow_ipv4_filter val; 1956 struct ib_flow_ipv4_filter mask; 1957 }; 1958 1959 struct ib_flow_ipv6_filter { 1960 u8 src_ip[16]; 1961 u8 dst_ip[16]; 1962 __be32 flow_label; 1963 u8 next_hdr; 1964 u8 traffic_class; 1965 u8 hop_limit; 1966 /* Must be last */ 1967 u8 real_sz[0]; 1968 }; 1969 1970 struct ib_flow_spec_ipv6 { 1971 u32 type; 1972 u16 size; 1973 struct ib_flow_ipv6_filter val; 1974 struct ib_flow_ipv6_filter mask; 1975 }; 1976 1977 struct ib_flow_tcp_udp_filter { 1978 __be16 dst_port; 1979 __be16 src_port; 1980 /* Must be last */ 1981 u8 real_sz[0]; 1982 }; 1983 1984 struct ib_flow_spec_tcp_udp { 1985 u32 type; 1986 u16 size; 1987 struct ib_flow_tcp_udp_filter val; 1988 struct ib_flow_tcp_udp_filter mask; 1989 }; 1990 1991 struct ib_flow_tunnel_filter { 1992 __be32 tunnel_id; 1993 u8 real_sz[0]; 1994 }; 1995 1996 /* ib_flow_spec_tunnel describes the Vxlan tunnel 1997 * the tunnel_id from val has the vni value 1998 */ 1999 struct ib_flow_spec_tunnel { 2000 u32 type; 2001 u16 size; 2002 struct ib_flow_tunnel_filter val; 2003 struct ib_flow_tunnel_filter mask; 2004 }; 2005 2006 struct ib_flow_esp_filter { 2007 __be32 spi; 2008 __be32 seq; 2009 /* Must be last */ 2010 u8 real_sz[0]; 2011 }; 2012 2013 struct ib_flow_spec_esp { 2014 u32 type; 2015 u16 size; 2016 struct ib_flow_esp_filter val; 2017 struct ib_flow_esp_filter mask; 2018 }; 2019 2020 struct ib_flow_gre_filter { 2021 __be16 c_ks_res0_ver; 2022 __be16 protocol; 2023 __be32 key; 2024 /* Must be last */ 2025 u8 real_sz[0]; 2026 }; 2027 2028 struct ib_flow_spec_gre { 2029 u32 type; 2030 u16 size; 2031 struct ib_flow_gre_filter val; 2032 struct ib_flow_gre_filter mask; 2033 }; 2034 2035 struct ib_flow_mpls_filter { 2036 __be32 tag; 2037 /* Must be last */ 2038 u8 real_sz[0]; 2039 }; 2040 2041 struct ib_flow_spec_mpls { 2042 u32 type; 2043 u16 size; 2044 struct ib_flow_mpls_filter val; 2045 struct ib_flow_mpls_filter mask; 2046 }; 2047 2048 struct ib_flow_spec_action_tag { 2049 enum ib_flow_spec_type type; 2050 u16 size; 2051 u32 tag_id; 2052 }; 2053 2054 struct ib_flow_spec_action_drop { 2055 enum ib_flow_spec_type type; 2056 u16 size; 2057 }; 2058 2059 struct ib_flow_spec_action_handle { 2060 enum ib_flow_spec_type type; 2061 u16 size; 2062 struct ib_flow_action *act; 2063 }; 2064 2065 enum ib_counters_description { 2066 IB_COUNTER_PACKETS, 2067 IB_COUNTER_BYTES, 2068 }; 2069 2070 struct ib_flow_spec_action_count { 2071 enum ib_flow_spec_type type; 2072 u16 size; 2073 struct ib_counters *counters; 2074 }; 2075 2076 union ib_flow_spec { 2077 struct { 2078 u32 type; 2079 u16 size; 2080 }; 2081 struct ib_flow_spec_eth eth; 2082 struct ib_flow_spec_ib ib; 2083 struct ib_flow_spec_ipv4 ipv4; 2084 struct ib_flow_spec_tcp_udp tcp_udp; 2085 struct ib_flow_spec_ipv6 ipv6; 2086 struct ib_flow_spec_tunnel tunnel; 2087 struct ib_flow_spec_esp esp; 2088 struct ib_flow_spec_gre gre; 2089 struct ib_flow_spec_mpls mpls; 2090 struct ib_flow_spec_action_tag flow_tag; 2091 struct ib_flow_spec_action_drop drop; 2092 struct ib_flow_spec_action_handle action; 2093 struct ib_flow_spec_action_count flow_count; 2094 }; 2095 2096 struct ib_flow_attr { 2097 enum ib_flow_attr_type type; 2098 u16 size; 2099 u16 priority; 2100 u32 flags; 2101 u8 num_of_specs; 2102 u8 port; 2103 union ib_flow_spec flows[]; 2104 }; 2105 2106 struct ib_flow { 2107 struct ib_qp *qp; 2108 struct ib_uobject *uobject; 2109 }; 2110 2111 enum ib_flow_action_type { 2112 IB_FLOW_ACTION_UNSPECIFIED, 2113 IB_FLOW_ACTION_ESP = 1, 2114 }; 2115 2116 struct ib_flow_action_attrs_esp_keymats { 2117 enum ib_uverbs_flow_action_esp_keymat protocol; 2118 union { 2119 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm; 2120 } keymat; 2121 }; 2122 2123 struct ib_flow_action_attrs_esp_replays { 2124 enum ib_uverbs_flow_action_esp_replay protocol; 2125 union { 2126 struct ib_uverbs_flow_action_esp_replay_bmp bmp; 2127 } replay; 2128 }; 2129 2130 enum ib_flow_action_attrs_esp_flags { 2131 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags 2132 * This is done in order to share the same flags between user-space and 2133 * kernel and spare an unnecessary translation. 2134 */ 2135 2136 /* Kernel flags */ 2137 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32, 2138 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33, 2139 }; 2140 2141 struct ib_flow_spec_list { 2142 struct ib_flow_spec_list *next; 2143 union ib_flow_spec spec; 2144 }; 2145 2146 struct ib_flow_action_attrs_esp { 2147 struct ib_flow_action_attrs_esp_keymats *keymat; 2148 struct ib_flow_action_attrs_esp_replays *replay; 2149 struct ib_flow_spec_list *encap; 2150 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled. 2151 * Value of 0 is a valid value. 2152 */ 2153 u32 esn; 2154 u32 spi; 2155 u32 seq; 2156 u32 tfc_pad; 2157 /* Use enum ib_flow_action_attrs_esp_flags */ 2158 u64 flags; 2159 u64 hard_limit_pkts; 2160 }; 2161 2162 struct ib_flow_action { 2163 struct ib_device *device; 2164 struct ib_uobject *uobject; 2165 enum ib_flow_action_type type; 2166 atomic_t usecnt; 2167 }; 2168 2169 struct ib_mad_hdr; 2170 struct ib_grh; 2171 2172 enum ib_process_mad_flags { 2173 IB_MAD_IGNORE_MKEY = 1, 2174 IB_MAD_IGNORE_BKEY = 2, 2175 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY 2176 }; 2177 2178 enum ib_mad_result { 2179 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */ 2180 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */ 2181 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */ 2182 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */ 2183 }; 2184 2185 struct ib_port_cache { 2186 u64 subnet_prefix; 2187 struct ib_pkey_cache *pkey; 2188 struct ib_gid_table *gid; 2189 u8 lmc; 2190 enum ib_port_state port_state; 2191 }; 2192 2193 struct ib_cache { 2194 rwlock_t lock; 2195 struct ib_event_handler event_handler; 2196 struct ib_port_cache *ports; 2197 }; 2198 2199 struct iw_cm_verbs; 2200 2201 struct ib_port_immutable { 2202 int pkey_tbl_len; 2203 int gid_tbl_len; 2204 u32 core_cap_flags; 2205 u32 max_mad_size; 2206 }; 2207 2208 /* rdma netdev type - specifies protocol type */ 2209 enum rdma_netdev_t { 2210 RDMA_NETDEV_OPA_VNIC, 2211 RDMA_NETDEV_IPOIB, 2212 }; 2213 2214 /** 2215 * struct rdma_netdev - rdma netdev 2216 * For cases where netstack interfacing is required. 2217 */ 2218 struct rdma_netdev { 2219 void *clnt_priv; 2220 struct ib_device *hca; 2221 u8 port_num; 2222 2223 /* cleanup function must be specified */ 2224 void (*free_rdma_netdev)(struct net_device *netdev); 2225 2226 /* control functions */ 2227 void (*set_id)(struct net_device *netdev, int id); 2228 /* send packet */ 2229 int (*send)(struct net_device *dev, struct sk_buff *skb, 2230 struct ib_ah *address, u32 dqpn); 2231 /* multicast */ 2232 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca, 2233 union ib_gid *gid, u16 mlid, 2234 int set_qkey, u32 qkey); 2235 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca, 2236 union ib_gid *gid, u16 mlid); 2237 }; 2238 2239 struct ib_port_pkey_list { 2240 /* Lock to hold while modifying the list. */ 2241 spinlock_t list_lock; 2242 struct list_head pkey_list; 2243 }; 2244 2245 struct ib_counters { 2246 struct ib_device *device; 2247 struct ib_uobject *uobject; 2248 /* num of objects attached */ 2249 atomic_t usecnt; 2250 }; 2251 2252 enum ib_read_counters_flags { 2253 /* prefer read values from driver cache */ 2254 IB_READ_COUNTERS_ATTR_PREFER_CACHED = 1 << 0, 2255 }; 2256 2257 struct ib_counters_read_attr { 2258 u64 *counters_buff; 2259 u32 ncounters; 2260 u32 flags; /* use enum ib_read_counters_flags */ 2261 }; 2262 2263 struct uverbs_attr_bundle; 2264 2265 struct ib_device { 2266 /* Do not access @dma_device directly from ULP nor from HW drivers. */ 2267 struct device *dma_device; 2268 2269 char name[IB_DEVICE_NAME_MAX]; 2270 2271 struct list_head event_handler_list; 2272 spinlock_t event_handler_lock; 2273 2274 spinlock_t client_data_lock; 2275 struct list_head core_list; 2276 /* Access to the client_data_list is protected by the client_data_lock 2277 * spinlock and the lists_rwsem read-write semaphore */ 2278 struct list_head client_data_list; 2279 2280 struct ib_cache cache; 2281 /** 2282 * port_immutable is indexed by port number 2283 */ 2284 struct ib_port_immutable *port_immutable; 2285 2286 int num_comp_vectors; 2287 2288 struct ib_port_pkey_list *port_pkey_list; 2289 2290 struct iw_cm_verbs *iwcm; 2291 2292 /** 2293 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the 2294 * driver initialized data. The struct is kfree()'ed by the sysfs 2295 * core when the device is removed. A lifespan of -1 in the return 2296 * struct tells the core to set a default lifespan. 2297 */ 2298 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device, 2299 u8 port_num); 2300 /** 2301 * get_hw_stats - Fill in the counter value(s) in the stats struct. 2302 * @index - The index in the value array we wish to have updated, or 2303 * num_counters if we want all stats updated 2304 * Return codes - 2305 * < 0 - Error, no counters updated 2306 * index - Updated the single counter pointed to by index 2307 * num_counters - Updated all counters (will reset the timestamp 2308 * and prevent further calls for lifespan milliseconds) 2309 * Drivers are allowed to update all counters in leiu of just the 2310 * one given in index at their option 2311 */ 2312 int (*get_hw_stats)(struct ib_device *device, 2313 struct rdma_hw_stats *stats, 2314 u8 port, int index); 2315 int (*query_device)(struct ib_device *device, 2316 struct ib_device_attr *device_attr, 2317 struct ib_udata *udata); 2318 int (*query_port)(struct ib_device *device, 2319 u8 port_num, 2320 struct ib_port_attr *port_attr); 2321 enum rdma_link_layer (*get_link_layer)(struct ib_device *device, 2322 u8 port_num); 2323 /* When calling get_netdev, the HW vendor's driver should return the 2324 * net device of device @device at port @port_num or NULL if such 2325 * a net device doesn't exist. The vendor driver should call dev_hold 2326 * on this net device. The HW vendor's device driver must guarantee 2327 * that this function returns NULL before the net device has finished 2328 * NETDEV_UNREGISTER state. 2329 */ 2330 struct net_device *(*get_netdev)(struct ib_device *device, 2331 u8 port_num); 2332 /* query_gid should be return GID value for @device, when @port_num 2333 * link layer is either IB or iWarp. It is no-op if @port_num port 2334 * is RoCE link layer. 2335 */ 2336 int (*query_gid)(struct ib_device *device, 2337 u8 port_num, int index, 2338 union ib_gid *gid); 2339 /* When calling add_gid, the HW vendor's driver should add the gid 2340 * of device of port at gid index available at @attr. Meta-info of 2341 * that gid (for example, the network device related to this gid) is 2342 * available at @attr. @context allows the HW vendor driver to store 2343 * extra information together with a GID entry. The HW vendor driver may 2344 * allocate memory to contain this information and store it in @context 2345 * when a new GID entry is written to. Params are consistent until the 2346 * next call of add_gid or delete_gid. The function should return 0 on 2347 * success or error otherwise. The function could be called 2348 * concurrently for different ports. This function is only called when 2349 * roce_gid_table is used. 2350 */ 2351 int (*add_gid)(const struct ib_gid_attr *attr, 2352 void **context); 2353 /* When calling del_gid, the HW vendor's driver should delete the 2354 * gid of device @device at gid index gid_index of port port_num 2355 * available in @attr. 2356 * Upon the deletion of a GID entry, the HW vendor must free any 2357 * allocated memory. The caller will clear @context afterwards. 2358 * This function is only called when roce_gid_table is used. 2359 */ 2360 int (*del_gid)(const struct ib_gid_attr *attr, 2361 void **context); 2362 int (*query_pkey)(struct ib_device *device, 2363 u8 port_num, u16 index, u16 *pkey); 2364 int (*modify_device)(struct ib_device *device, 2365 int device_modify_mask, 2366 struct ib_device_modify *device_modify); 2367 int (*modify_port)(struct ib_device *device, 2368 u8 port_num, int port_modify_mask, 2369 struct ib_port_modify *port_modify); 2370 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device, 2371 struct ib_udata *udata); 2372 int (*dealloc_ucontext)(struct ib_ucontext *context); 2373 int (*mmap)(struct ib_ucontext *context, 2374 struct vm_area_struct *vma); 2375 struct ib_pd * (*alloc_pd)(struct ib_device *device, 2376 struct ib_ucontext *context, 2377 struct ib_udata *udata); 2378 int (*dealloc_pd)(struct ib_pd *pd); 2379 struct ib_ah * (*create_ah)(struct ib_pd *pd, 2380 struct rdma_ah_attr *ah_attr, 2381 struct ib_udata *udata); 2382 int (*modify_ah)(struct ib_ah *ah, 2383 struct rdma_ah_attr *ah_attr); 2384 int (*query_ah)(struct ib_ah *ah, 2385 struct rdma_ah_attr *ah_attr); 2386 int (*destroy_ah)(struct ib_ah *ah); 2387 struct ib_srq * (*create_srq)(struct ib_pd *pd, 2388 struct ib_srq_init_attr *srq_init_attr, 2389 struct ib_udata *udata); 2390 int (*modify_srq)(struct ib_srq *srq, 2391 struct ib_srq_attr *srq_attr, 2392 enum ib_srq_attr_mask srq_attr_mask, 2393 struct ib_udata *udata); 2394 int (*query_srq)(struct ib_srq *srq, 2395 struct ib_srq_attr *srq_attr); 2396 int (*destroy_srq)(struct ib_srq *srq); 2397 int (*post_srq_recv)(struct ib_srq *srq, 2398 struct ib_recv_wr *recv_wr, 2399 struct ib_recv_wr **bad_recv_wr); 2400 struct ib_qp * (*create_qp)(struct ib_pd *pd, 2401 struct ib_qp_init_attr *qp_init_attr, 2402 struct ib_udata *udata); 2403 int (*modify_qp)(struct ib_qp *qp, 2404 struct ib_qp_attr *qp_attr, 2405 int qp_attr_mask, 2406 struct ib_udata *udata); 2407 int (*query_qp)(struct ib_qp *qp, 2408 struct ib_qp_attr *qp_attr, 2409 int qp_attr_mask, 2410 struct ib_qp_init_attr *qp_init_attr); 2411 int (*destroy_qp)(struct ib_qp *qp); 2412 int (*post_send)(struct ib_qp *qp, 2413 struct ib_send_wr *send_wr, 2414 struct ib_send_wr **bad_send_wr); 2415 int (*post_recv)(struct ib_qp *qp, 2416 struct ib_recv_wr *recv_wr, 2417 struct ib_recv_wr **bad_recv_wr); 2418 struct ib_cq * (*create_cq)(struct ib_device *device, 2419 const struct ib_cq_init_attr *attr, 2420 struct ib_ucontext *context, 2421 struct ib_udata *udata); 2422 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, 2423 u16 cq_period); 2424 int (*destroy_cq)(struct ib_cq *cq); 2425 int (*resize_cq)(struct ib_cq *cq, int cqe, 2426 struct ib_udata *udata); 2427 int (*poll_cq)(struct ib_cq *cq, int num_entries, 2428 struct ib_wc *wc); 2429 int (*peek_cq)(struct ib_cq *cq, int wc_cnt); 2430 int (*req_notify_cq)(struct ib_cq *cq, 2431 enum ib_cq_notify_flags flags); 2432 int (*req_ncomp_notif)(struct ib_cq *cq, 2433 int wc_cnt); 2434 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd, 2435 int mr_access_flags); 2436 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd, 2437 u64 start, u64 length, 2438 u64 virt_addr, 2439 int mr_access_flags, 2440 struct ib_udata *udata); 2441 int (*rereg_user_mr)(struct ib_mr *mr, 2442 int flags, 2443 u64 start, u64 length, 2444 u64 virt_addr, 2445 int mr_access_flags, 2446 struct ib_pd *pd, 2447 struct ib_udata *udata); 2448 int (*dereg_mr)(struct ib_mr *mr); 2449 struct ib_mr * (*alloc_mr)(struct ib_pd *pd, 2450 enum ib_mr_type mr_type, 2451 u32 max_num_sg); 2452 int (*map_mr_sg)(struct ib_mr *mr, 2453 struct scatterlist *sg, 2454 int sg_nents, 2455 unsigned int *sg_offset); 2456 struct ib_mw * (*alloc_mw)(struct ib_pd *pd, 2457 enum ib_mw_type type, 2458 struct ib_udata *udata); 2459 int (*dealloc_mw)(struct ib_mw *mw); 2460 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd, 2461 int mr_access_flags, 2462 struct ib_fmr_attr *fmr_attr); 2463 int (*map_phys_fmr)(struct ib_fmr *fmr, 2464 u64 *page_list, int list_len, 2465 u64 iova); 2466 int (*unmap_fmr)(struct list_head *fmr_list); 2467 int (*dealloc_fmr)(struct ib_fmr *fmr); 2468 int (*attach_mcast)(struct ib_qp *qp, 2469 union ib_gid *gid, 2470 u16 lid); 2471 int (*detach_mcast)(struct ib_qp *qp, 2472 union ib_gid *gid, 2473 u16 lid); 2474 int (*process_mad)(struct ib_device *device, 2475 int process_mad_flags, 2476 u8 port_num, 2477 const struct ib_wc *in_wc, 2478 const struct ib_grh *in_grh, 2479 const struct ib_mad_hdr *in_mad, 2480 size_t in_mad_size, 2481 struct ib_mad_hdr *out_mad, 2482 size_t *out_mad_size, 2483 u16 *out_mad_pkey_index); 2484 struct ib_xrcd * (*alloc_xrcd)(struct ib_device *device, 2485 struct ib_ucontext *ucontext, 2486 struct ib_udata *udata); 2487 int (*dealloc_xrcd)(struct ib_xrcd *xrcd); 2488 struct ib_flow * (*create_flow)(struct ib_qp *qp, 2489 struct ib_flow_attr 2490 *flow_attr, 2491 int domain, 2492 struct ib_udata *udata); 2493 int (*destroy_flow)(struct ib_flow *flow_id); 2494 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask, 2495 struct ib_mr_status *mr_status); 2496 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext); 2497 void (*drain_rq)(struct ib_qp *qp); 2498 void (*drain_sq)(struct ib_qp *qp); 2499 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port, 2500 int state); 2501 int (*get_vf_config)(struct ib_device *device, int vf, u8 port, 2502 struct ifla_vf_info *ivf); 2503 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port, 2504 struct ifla_vf_stats *stats); 2505 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid, 2506 int type); 2507 struct ib_wq * (*create_wq)(struct ib_pd *pd, 2508 struct ib_wq_init_attr *init_attr, 2509 struct ib_udata *udata); 2510 int (*destroy_wq)(struct ib_wq *wq); 2511 int (*modify_wq)(struct ib_wq *wq, 2512 struct ib_wq_attr *attr, 2513 u32 wq_attr_mask, 2514 struct ib_udata *udata); 2515 struct ib_rwq_ind_table * (*create_rwq_ind_table)(struct ib_device *device, 2516 struct ib_rwq_ind_table_init_attr *init_attr, 2517 struct ib_udata *udata); 2518 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table); 2519 struct ib_flow_action * (*create_flow_action_esp)(struct ib_device *device, 2520 const struct ib_flow_action_attrs_esp *attr, 2521 struct uverbs_attr_bundle *attrs); 2522 int (*destroy_flow_action)(struct ib_flow_action *action); 2523 int (*modify_flow_action_esp)(struct ib_flow_action *action, 2524 const struct ib_flow_action_attrs_esp *attr, 2525 struct uverbs_attr_bundle *attrs); 2526 struct ib_dm * (*alloc_dm)(struct ib_device *device, 2527 struct ib_ucontext *context, 2528 struct ib_dm_alloc_attr *attr, 2529 struct uverbs_attr_bundle *attrs); 2530 int (*dealloc_dm)(struct ib_dm *dm); 2531 struct ib_mr * (*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm, 2532 struct ib_dm_mr_attr *attr, 2533 struct uverbs_attr_bundle *attrs); 2534 struct ib_counters * (*create_counters)(struct ib_device *device, 2535 struct uverbs_attr_bundle *attrs); 2536 int (*destroy_counters)(struct ib_counters *counters); 2537 int (*read_counters)(struct ib_counters *counters, 2538 struct ib_counters_read_attr *counters_read_attr, 2539 struct uverbs_attr_bundle *attrs); 2540 2541 /** 2542 * rdma netdev operation 2543 * 2544 * Driver implementing alloc_rdma_netdev must return -EOPNOTSUPP if it 2545 * doesn't support the specified rdma netdev type. 2546 */ 2547 struct net_device *(*alloc_rdma_netdev)( 2548 struct ib_device *device, 2549 u8 port_num, 2550 enum rdma_netdev_t type, 2551 const char *name, 2552 unsigned char name_assign_type, 2553 void (*setup)(struct net_device *)); 2554 2555 struct module *owner; 2556 struct device dev; 2557 struct kobject *ports_parent; 2558 struct list_head port_list; 2559 2560 enum { 2561 IB_DEV_UNINITIALIZED, 2562 IB_DEV_REGISTERED, 2563 IB_DEV_UNREGISTERED 2564 } reg_state; 2565 2566 int uverbs_abi_ver; 2567 u64 uverbs_cmd_mask; 2568 u64 uverbs_ex_cmd_mask; 2569 2570 char node_desc[IB_DEVICE_NODE_DESC_MAX]; 2571 __be64 node_guid; 2572 u32 local_dma_lkey; 2573 u16 is_switch:1; 2574 u8 node_type; 2575 u8 phys_port_cnt; 2576 struct ib_device_attr attrs; 2577 struct attribute_group *hw_stats_ag; 2578 struct rdma_hw_stats *hw_stats; 2579 2580 #ifdef CONFIG_CGROUP_RDMA 2581 struct rdmacg_device cg_device; 2582 #endif 2583 2584 u32 index; 2585 /* 2586 * Implementation details of the RDMA core, don't use in drivers 2587 */ 2588 struct rdma_restrack_root res; 2589 2590 /** 2591 * The following mandatory functions are used only at device 2592 * registration. Keep functions such as these at the end of this 2593 * structure to avoid cache line misses when accessing struct ib_device 2594 * in fast paths. 2595 */ 2596 int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *); 2597 void (*get_dev_fw_str)(struct ib_device *, char *str); 2598 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev, 2599 int comp_vector); 2600 2601 struct uverbs_root_spec *specs_root; 2602 enum rdma_driver_id driver_id; 2603 }; 2604 2605 struct ib_client { 2606 char *name; 2607 void (*add) (struct ib_device *); 2608 void (*remove)(struct ib_device *, void *client_data); 2609 2610 /* Returns the net_dev belonging to this ib_client and matching the 2611 * given parameters. 2612 * @dev: An RDMA device that the net_dev use for communication. 2613 * @port: A physical port number on the RDMA device. 2614 * @pkey: P_Key that the net_dev uses if applicable. 2615 * @gid: A GID that the net_dev uses to communicate. 2616 * @addr: An IP address the net_dev is configured with. 2617 * @client_data: The device's client data set by ib_set_client_data(). 2618 * 2619 * An ib_client that implements a net_dev on top of RDMA devices 2620 * (such as IP over IB) should implement this callback, allowing the 2621 * rdma_cm module to find the right net_dev for a given request. 2622 * 2623 * The caller is responsible for calling dev_put on the returned 2624 * netdev. */ 2625 struct net_device *(*get_net_dev_by_params)( 2626 struct ib_device *dev, 2627 u8 port, 2628 u16 pkey, 2629 const union ib_gid *gid, 2630 const struct sockaddr *addr, 2631 void *client_data); 2632 struct list_head list; 2633 }; 2634 2635 struct ib_device *ib_alloc_device(size_t size); 2636 void ib_dealloc_device(struct ib_device *device); 2637 2638 void ib_get_device_fw_str(struct ib_device *device, char *str); 2639 2640 int ib_register_device(struct ib_device *device, 2641 int (*port_callback)(struct ib_device *, 2642 u8, struct kobject *)); 2643 void ib_unregister_device(struct ib_device *device); 2644 2645 int ib_register_client (struct ib_client *client); 2646 void ib_unregister_client(struct ib_client *client); 2647 2648 void *ib_get_client_data(struct ib_device *device, struct ib_client *client); 2649 void ib_set_client_data(struct ib_device *device, struct ib_client *client, 2650 void *data); 2651 2652 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len) 2653 { 2654 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0; 2655 } 2656 2657 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len) 2658 { 2659 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0; 2660 } 2661 2662 static inline bool ib_is_buffer_cleared(const void __user *p, 2663 size_t len) 2664 { 2665 bool ret; 2666 u8 *buf; 2667 2668 if (len > USHRT_MAX) 2669 return false; 2670 2671 buf = memdup_user(p, len); 2672 if (IS_ERR(buf)) 2673 return false; 2674 2675 ret = !memchr_inv(buf, 0, len); 2676 kfree(buf); 2677 return ret; 2678 } 2679 2680 static inline bool ib_is_udata_cleared(struct ib_udata *udata, 2681 size_t offset, 2682 size_t len) 2683 { 2684 return ib_is_buffer_cleared(udata->inbuf + offset, len); 2685 } 2686 2687 /** 2688 * ib_modify_qp_is_ok - Check that the supplied attribute mask 2689 * contains all required attributes and no attributes not allowed for 2690 * the given QP state transition. 2691 * @cur_state: Current QP state 2692 * @next_state: Next QP state 2693 * @type: QP type 2694 * @mask: Mask of supplied QP attributes 2695 * @ll : link layer of port 2696 * 2697 * This function is a helper function that a low-level driver's 2698 * modify_qp method can use to validate the consumer's input. It 2699 * checks that cur_state and next_state are valid QP states, that a 2700 * transition from cur_state to next_state is allowed by the IB spec, 2701 * and that the attribute mask supplied is allowed for the transition. 2702 */ 2703 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state, 2704 enum ib_qp_type type, enum ib_qp_attr_mask mask, 2705 enum rdma_link_layer ll); 2706 2707 void ib_register_event_handler(struct ib_event_handler *event_handler); 2708 void ib_unregister_event_handler(struct ib_event_handler *event_handler); 2709 void ib_dispatch_event(struct ib_event *event); 2710 2711 int ib_query_port(struct ib_device *device, 2712 u8 port_num, struct ib_port_attr *port_attr); 2713 2714 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device, 2715 u8 port_num); 2716 2717 /** 2718 * rdma_cap_ib_switch - Check if the device is IB switch 2719 * @device: Device to check 2720 * 2721 * Device driver is responsible for setting is_switch bit on 2722 * in ib_device structure at init time. 2723 * 2724 * Return: true if the device is IB switch. 2725 */ 2726 static inline bool rdma_cap_ib_switch(const struct ib_device *device) 2727 { 2728 return device->is_switch; 2729 } 2730 2731 /** 2732 * rdma_start_port - Return the first valid port number for the device 2733 * specified 2734 * 2735 * @device: Device to be checked 2736 * 2737 * Return start port number 2738 */ 2739 static inline u8 rdma_start_port(const struct ib_device *device) 2740 { 2741 return rdma_cap_ib_switch(device) ? 0 : 1; 2742 } 2743 2744 /** 2745 * rdma_end_port - Return the last valid port number for the device 2746 * specified 2747 * 2748 * @device: Device to be checked 2749 * 2750 * Return last port number 2751 */ 2752 static inline u8 rdma_end_port(const struct ib_device *device) 2753 { 2754 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt; 2755 } 2756 2757 static inline int rdma_is_port_valid(const struct ib_device *device, 2758 unsigned int port) 2759 { 2760 return (port >= rdma_start_port(device) && 2761 port <= rdma_end_port(device)); 2762 } 2763 2764 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num) 2765 { 2766 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB; 2767 } 2768 2769 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num) 2770 { 2771 return device->port_immutable[port_num].core_cap_flags & 2772 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP); 2773 } 2774 2775 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num) 2776 { 2777 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP; 2778 } 2779 2780 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num) 2781 { 2782 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE; 2783 } 2784 2785 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num) 2786 { 2787 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP; 2788 } 2789 2790 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num) 2791 { 2792 return rdma_protocol_ib(device, port_num) || 2793 rdma_protocol_roce(device, port_num); 2794 } 2795 2796 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num) 2797 { 2798 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_RAW_PACKET; 2799 } 2800 2801 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num) 2802 { 2803 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_USNIC; 2804 } 2805 2806 /** 2807 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband 2808 * Management Datagrams. 2809 * @device: Device to check 2810 * @port_num: Port number to check 2811 * 2812 * Management Datagrams (MAD) are a required part of the InfiniBand 2813 * specification and are supported on all InfiniBand devices. A slightly 2814 * extended version are also supported on OPA interfaces. 2815 * 2816 * Return: true if the port supports sending/receiving of MAD packets. 2817 */ 2818 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num) 2819 { 2820 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD; 2821 } 2822 2823 /** 2824 * rdma_cap_opa_mad - Check if the port of device provides support for OPA 2825 * Management Datagrams. 2826 * @device: Device to check 2827 * @port_num: Port number to check 2828 * 2829 * Intel OmniPath devices extend and/or replace the InfiniBand Management 2830 * datagrams with their own versions. These OPA MADs share many but not all of 2831 * the characteristics of InfiniBand MADs. 2832 * 2833 * OPA MADs differ in the following ways: 2834 * 2835 * 1) MADs are variable size up to 2K 2836 * IBTA defined MADs remain fixed at 256 bytes 2837 * 2) OPA SMPs must carry valid PKeys 2838 * 3) OPA SMP packets are a different format 2839 * 2840 * Return: true if the port supports OPA MAD packet formats. 2841 */ 2842 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num) 2843 { 2844 return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD) 2845 == RDMA_CORE_CAP_OPA_MAD; 2846 } 2847 2848 /** 2849 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband 2850 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI). 2851 * @device: Device to check 2852 * @port_num: Port number to check 2853 * 2854 * Each InfiniBand node is required to provide a Subnet Management Agent 2855 * that the subnet manager can access. Prior to the fabric being fully 2856 * configured by the subnet manager, the SMA is accessed via a well known 2857 * interface called the Subnet Management Interface (SMI). This interface 2858 * uses directed route packets to communicate with the SM to get around the 2859 * chicken and egg problem of the SM needing to know what's on the fabric 2860 * in order to configure the fabric, and needing to configure the fabric in 2861 * order to send packets to the devices on the fabric. These directed 2862 * route packets do not need the fabric fully configured in order to reach 2863 * their destination. The SMI is the only method allowed to send 2864 * directed route packets on an InfiniBand fabric. 2865 * 2866 * Return: true if the port provides an SMI. 2867 */ 2868 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num) 2869 { 2870 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI; 2871 } 2872 2873 /** 2874 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband 2875 * Communication Manager. 2876 * @device: Device to check 2877 * @port_num: Port number to check 2878 * 2879 * The InfiniBand Communication Manager is one of many pre-defined General 2880 * Service Agents (GSA) that are accessed via the General Service 2881 * Interface (GSI). It's role is to facilitate establishment of connections 2882 * between nodes as well as other management related tasks for established 2883 * connections. 2884 * 2885 * Return: true if the port supports an IB CM (this does not guarantee that 2886 * a CM is actually running however). 2887 */ 2888 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num) 2889 { 2890 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM; 2891 } 2892 2893 /** 2894 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP 2895 * Communication Manager. 2896 * @device: Device to check 2897 * @port_num: Port number to check 2898 * 2899 * Similar to above, but specific to iWARP connections which have a different 2900 * managment protocol than InfiniBand. 2901 * 2902 * Return: true if the port supports an iWARP CM (this does not guarantee that 2903 * a CM is actually running however). 2904 */ 2905 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num) 2906 { 2907 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM; 2908 } 2909 2910 /** 2911 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband 2912 * Subnet Administration. 2913 * @device: Device to check 2914 * @port_num: Port number to check 2915 * 2916 * An InfiniBand Subnet Administration (SA) service is a pre-defined General 2917 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand 2918 * fabrics, devices should resolve routes to other hosts by contacting the 2919 * SA to query the proper route. 2920 * 2921 * Return: true if the port should act as a client to the fabric Subnet 2922 * Administration interface. This does not imply that the SA service is 2923 * running locally. 2924 */ 2925 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num) 2926 { 2927 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA; 2928 } 2929 2930 /** 2931 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband 2932 * Multicast. 2933 * @device: Device to check 2934 * @port_num: Port number to check 2935 * 2936 * InfiniBand multicast registration is more complex than normal IPv4 or 2937 * IPv6 multicast registration. Each Host Channel Adapter must register 2938 * with the Subnet Manager when it wishes to join a multicast group. It 2939 * should do so only once regardless of how many queue pairs it subscribes 2940 * to this group. And it should leave the group only after all queue pairs 2941 * attached to the group have been detached. 2942 * 2943 * Return: true if the port must undertake the additional adminstrative 2944 * overhead of registering/unregistering with the SM and tracking of the 2945 * total number of queue pairs attached to the multicast group. 2946 */ 2947 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num) 2948 { 2949 return rdma_cap_ib_sa(device, port_num); 2950 } 2951 2952 /** 2953 * rdma_cap_af_ib - Check if the port of device has the capability 2954 * Native Infiniband Address. 2955 * @device: Device to check 2956 * @port_num: Port number to check 2957 * 2958 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default 2959 * GID. RoCE uses a different mechanism, but still generates a GID via 2960 * a prescribed mechanism and port specific data. 2961 * 2962 * Return: true if the port uses a GID address to identify devices on the 2963 * network. 2964 */ 2965 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num) 2966 { 2967 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB; 2968 } 2969 2970 /** 2971 * rdma_cap_eth_ah - Check if the port of device has the capability 2972 * Ethernet Address Handle. 2973 * @device: Device to check 2974 * @port_num: Port number to check 2975 * 2976 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique 2977 * to fabricate GIDs over Ethernet/IP specific addresses native to the 2978 * port. Normally, packet headers are generated by the sending host 2979 * adapter, but when sending connectionless datagrams, we must manually 2980 * inject the proper headers for the fabric we are communicating over. 2981 * 2982 * Return: true if we are running as a RoCE port and must force the 2983 * addition of a Global Route Header built from our Ethernet Address 2984 * Handle into our header list for connectionless packets. 2985 */ 2986 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num) 2987 { 2988 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH; 2989 } 2990 2991 /** 2992 * rdma_cap_opa_ah - Check if the port of device supports 2993 * OPA Address handles 2994 * @device: Device to check 2995 * @port_num: Port number to check 2996 * 2997 * Return: true if we are running on an OPA device which supports 2998 * the extended OPA addressing. 2999 */ 3000 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num) 3001 { 3002 return (device->port_immutable[port_num].core_cap_flags & 3003 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH; 3004 } 3005 3006 /** 3007 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port. 3008 * 3009 * @device: Device 3010 * @port_num: Port number 3011 * 3012 * This MAD size includes the MAD headers and MAD payload. No other headers 3013 * are included. 3014 * 3015 * Return the max MAD size required by the Port. Will return 0 if the port 3016 * does not support MADs 3017 */ 3018 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num) 3019 { 3020 return device->port_immutable[port_num].max_mad_size; 3021 } 3022 3023 /** 3024 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table 3025 * @device: Device to check 3026 * @port_num: Port number to check 3027 * 3028 * RoCE GID table mechanism manages the various GIDs for a device. 3029 * 3030 * NOTE: if allocating the port's GID table has failed, this call will still 3031 * return true, but any RoCE GID table API will fail. 3032 * 3033 * Return: true if the port uses RoCE GID table mechanism in order to manage 3034 * its GIDs. 3035 */ 3036 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device, 3037 u8 port_num) 3038 { 3039 return rdma_protocol_roce(device, port_num) && 3040 device->add_gid && device->del_gid; 3041 } 3042 3043 /* 3044 * Check if the device supports READ W/ INVALIDATE. 3045 */ 3046 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num) 3047 { 3048 /* 3049 * iWarp drivers must support READ W/ INVALIDATE. No other protocol 3050 * has support for it yet. 3051 */ 3052 return rdma_protocol_iwarp(dev, port_num); 3053 } 3054 3055 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port, 3056 int state); 3057 int ib_get_vf_config(struct ib_device *device, int vf, u8 port, 3058 struct ifla_vf_info *info); 3059 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port, 3060 struct ifla_vf_stats *stats); 3061 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid, 3062 int type); 3063 3064 int ib_query_pkey(struct ib_device *device, 3065 u8 port_num, u16 index, u16 *pkey); 3066 3067 int ib_modify_device(struct ib_device *device, 3068 int device_modify_mask, 3069 struct ib_device_modify *device_modify); 3070 3071 int ib_modify_port(struct ib_device *device, 3072 u8 port_num, int port_modify_mask, 3073 struct ib_port_modify *port_modify); 3074 3075 int ib_find_gid(struct ib_device *device, union ib_gid *gid, 3076 u8 *port_num, u16 *index); 3077 3078 int ib_find_pkey(struct ib_device *device, 3079 u8 port_num, u16 pkey, u16 *index); 3080 3081 enum ib_pd_flags { 3082 /* 3083 * Create a memory registration for all memory in the system and place 3084 * the rkey for it into pd->unsafe_global_rkey. This can be used by 3085 * ULPs to avoid the overhead of dynamic MRs. 3086 * 3087 * This flag is generally considered unsafe and must only be used in 3088 * extremly trusted environments. Every use of it will log a warning 3089 * in the kernel log. 3090 */ 3091 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01, 3092 }; 3093 3094 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags, 3095 const char *caller); 3096 #define ib_alloc_pd(device, flags) \ 3097 __ib_alloc_pd((device), (flags), KBUILD_MODNAME) 3098 void ib_dealloc_pd(struct ib_pd *pd); 3099 3100 /** 3101 * rdma_create_ah - Creates an address handle for the given address vector. 3102 * @pd: The protection domain associated with the address handle. 3103 * @ah_attr: The attributes of the address vector. 3104 * 3105 * The address handle is used to reference a local or global destination 3106 * in all UD QP post sends. 3107 */ 3108 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr); 3109 3110 /** 3111 * rdma_create_user_ah - Creates an address handle for the given address vector. 3112 * It resolves destination mac address for ah attribute of RoCE type. 3113 * @pd: The protection domain associated with the address handle. 3114 * @ah_attr: The attributes of the address vector. 3115 * @udata: pointer to user's input output buffer information need by 3116 * provider driver. 3117 * 3118 * It returns 0 on success and returns appropriate error code on error. 3119 * The address handle is used to reference a local or global destination 3120 * in all UD QP post sends. 3121 */ 3122 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd, 3123 struct rdma_ah_attr *ah_attr, 3124 struct ib_udata *udata); 3125 /** 3126 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header 3127 * work completion. 3128 * @hdr: the L3 header to parse 3129 * @net_type: type of header to parse 3130 * @sgid: place to store source gid 3131 * @dgid: place to store destination gid 3132 */ 3133 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr, 3134 enum rdma_network_type net_type, 3135 union ib_gid *sgid, union ib_gid *dgid); 3136 3137 /** 3138 * ib_get_rdma_header_version - Get the header version 3139 * @hdr: the L3 header to parse 3140 */ 3141 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr); 3142 3143 /** 3144 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a 3145 * work completion. 3146 * @device: Device on which the received message arrived. 3147 * @port_num: Port on which the received message arrived. 3148 * @wc: Work completion associated with the received message. 3149 * @grh: References the received global route header. This parameter is 3150 * ignored unless the work completion indicates that the GRH is valid. 3151 * @ah_attr: Returned attributes that can be used when creating an address 3152 * handle for replying to the message. 3153 */ 3154 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num, 3155 const struct ib_wc *wc, const struct ib_grh *grh, 3156 struct rdma_ah_attr *ah_attr); 3157 3158 /** 3159 * ib_create_ah_from_wc - Creates an address handle associated with the 3160 * sender of the specified work completion. 3161 * @pd: The protection domain associated with the address handle. 3162 * @wc: Work completion information associated with a received message. 3163 * @grh: References the received global route header. This parameter is 3164 * ignored unless the work completion indicates that the GRH is valid. 3165 * @port_num: The outbound port number to associate with the address. 3166 * 3167 * The address handle is used to reference a local or global destination 3168 * in all UD QP post sends. 3169 */ 3170 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc, 3171 const struct ib_grh *grh, u8 port_num); 3172 3173 /** 3174 * rdma_modify_ah - Modifies the address vector associated with an address 3175 * handle. 3176 * @ah: The address handle to modify. 3177 * @ah_attr: The new address vector attributes to associate with the 3178 * address handle. 3179 */ 3180 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr); 3181 3182 /** 3183 * rdma_query_ah - Queries the address vector associated with an address 3184 * handle. 3185 * @ah: The address handle to query. 3186 * @ah_attr: The address vector attributes associated with the address 3187 * handle. 3188 */ 3189 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr); 3190 3191 /** 3192 * rdma_destroy_ah - Destroys an address handle. 3193 * @ah: The address handle to destroy. 3194 */ 3195 int rdma_destroy_ah(struct ib_ah *ah); 3196 3197 /** 3198 * ib_create_srq - Creates a SRQ associated with the specified protection 3199 * domain. 3200 * @pd: The protection domain associated with the SRQ. 3201 * @srq_init_attr: A list of initial attributes required to create the 3202 * SRQ. If SRQ creation succeeds, then the attributes are updated to 3203 * the actual capabilities of the created SRQ. 3204 * 3205 * srq_attr->max_wr and srq_attr->max_sge are read the determine the 3206 * requested size of the SRQ, and set to the actual values allocated 3207 * on return. If ib_create_srq() succeeds, then max_wr and max_sge 3208 * will always be at least as large as the requested values. 3209 */ 3210 struct ib_srq *ib_create_srq(struct ib_pd *pd, 3211 struct ib_srq_init_attr *srq_init_attr); 3212 3213 /** 3214 * ib_modify_srq - Modifies the attributes for the specified SRQ. 3215 * @srq: The SRQ to modify. 3216 * @srq_attr: On input, specifies the SRQ attributes to modify. On output, 3217 * the current values of selected SRQ attributes are returned. 3218 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ 3219 * are being modified. 3220 * 3221 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or 3222 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when 3223 * the number of receives queued drops below the limit. 3224 */ 3225 int ib_modify_srq(struct ib_srq *srq, 3226 struct ib_srq_attr *srq_attr, 3227 enum ib_srq_attr_mask srq_attr_mask); 3228 3229 /** 3230 * ib_query_srq - Returns the attribute list and current values for the 3231 * specified SRQ. 3232 * @srq: The SRQ to query. 3233 * @srq_attr: The attributes of the specified SRQ. 3234 */ 3235 int ib_query_srq(struct ib_srq *srq, 3236 struct ib_srq_attr *srq_attr); 3237 3238 /** 3239 * ib_destroy_srq - Destroys the specified SRQ. 3240 * @srq: The SRQ to destroy. 3241 */ 3242 int ib_destroy_srq(struct ib_srq *srq); 3243 3244 /** 3245 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ. 3246 * @srq: The SRQ to post the work request on. 3247 * @recv_wr: A list of work requests to post on the receive queue. 3248 * @bad_recv_wr: On an immediate failure, this parameter will reference 3249 * the work request that failed to be posted on the QP. 3250 */ 3251 static inline int ib_post_srq_recv(struct ib_srq *srq, 3252 struct ib_recv_wr *recv_wr, 3253 struct ib_recv_wr **bad_recv_wr) 3254 { 3255 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr); 3256 } 3257 3258 /** 3259 * ib_create_qp - Creates a QP associated with the specified protection 3260 * domain. 3261 * @pd: The protection domain associated with the QP. 3262 * @qp_init_attr: A list of initial attributes required to create the 3263 * QP. If QP creation succeeds, then the attributes are updated to 3264 * the actual capabilities of the created QP. 3265 */ 3266 struct ib_qp *ib_create_qp(struct ib_pd *pd, 3267 struct ib_qp_init_attr *qp_init_attr); 3268 3269 /** 3270 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP. 3271 * @qp: The QP to modify. 3272 * @attr: On input, specifies the QP attributes to modify. On output, 3273 * the current values of selected QP attributes are returned. 3274 * @attr_mask: A bit-mask used to specify which attributes of the QP 3275 * are being modified. 3276 * @udata: pointer to user's input output buffer information 3277 * are being modified. 3278 * It returns 0 on success and returns appropriate error code on error. 3279 */ 3280 int ib_modify_qp_with_udata(struct ib_qp *qp, 3281 struct ib_qp_attr *attr, 3282 int attr_mask, 3283 struct ib_udata *udata); 3284 3285 /** 3286 * ib_modify_qp - Modifies the attributes for the specified QP and then 3287 * transitions the QP to the given state. 3288 * @qp: The QP to modify. 3289 * @qp_attr: On input, specifies the QP attributes to modify. On output, 3290 * the current values of selected QP attributes are returned. 3291 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP 3292 * are being modified. 3293 */ 3294 int ib_modify_qp(struct ib_qp *qp, 3295 struct ib_qp_attr *qp_attr, 3296 int qp_attr_mask); 3297 3298 /** 3299 * ib_query_qp - Returns the attribute list and current values for the 3300 * specified QP. 3301 * @qp: The QP to query. 3302 * @qp_attr: The attributes of the specified QP. 3303 * @qp_attr_mask: A bit-mask used to select specific attributes to query. 3304 * @qp_init_attr: Additional attributes of the selected QP. 3305 * 3306 * The qp_attr_mask may be used to limit the query to gathering only the 3307 * selected attributes. 3308 */ 3309 int ib_query_qp(struct ib_qp *qp, 3310 struct ib_qp_attr *qp_attr, 3311 int qp_attr_mask, 3312 struct ib_qp_init_attr *qp_init_attr); 3313 3314 /** 3315 * ib_destroy_qp - Destroys the specified QP. 3316 * @qp: The QP to destroy. 3317 */ 3318 int ib_destroy_qp(struct ib_qp *qp); 3319 3320 /** 3321 * ib_open_qp - Obtain a reference to an existing sharable QP. 3322 * @xrcd - XRC domain 3323 * @qp_open_attr: Attributes identifying the QP to open. 3324 * 3325 * Returns a reference to a sharable QP. 3326 */ 3327 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd, 3328 struct ib_qp_open_attr *qp_open_attr); 3329 3330 /** 3331 * ib_close_qp - Release an external reference to a QP. 3332 * @qp: The QP handle to release 3333 * 3334 * The opened QP handle is released by the caller. The underlying 3335 * shared QP is not destroyed until all internal references are released. 3336 */ 3337 int ib_close_qp(struct ib_qp *qp); 3338 3339 /** 3340 * ib_post_send - Posts a list of work requests to the send queue of 3341 * the specified QP. 3342 * @qp: The QP to post the work request on. 3343 * @send_wr: A list of work requests to post on the send queue. 3344 * @bad_send_wr: On an immediate failure, this parameter will reference 3345 * the work request that failed to be posted on the QP. 3346 * 3347 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate 3348 * error is returned, the QP state shall not be affected, 3349 * ib_post_send() will return an immediate error after queueing any 3350 * earlier work requests in the list. 3351 */ 3352 static inline int ib_post_send(struct ib_qp *qp, 3353 struct ib_send_wr *send_wr, 3354 struct ib_send_wr **bad_send_wr) 3355 { 3356 return qp->device->post_send(qp, send_wr, bad_send_wr); 3357 } 3358 3359 /** 3360 * ib_post_recv - Posts a list of work requests to the receive queue of 3361 * the specified QP. 3362 * @qp: The QP to post the work request on. 3363 * @recv_wr: A list of work requests to post on the receive queue. 3364 * @bad_recv_wr: On an immediate failure, this parameter will reference 3365 * the work request that failed to be posted on the QP. 3366 */ 3367 static inline int ib_post_recv(struct ib_qp *qp, 3368 struct ib_recv_wr *recv_wr, 3369 struct ib_recv_wr **bad_recv_wr) 3370 { 3371 return qp->device->post_recv(qp, recv_wr, bad_recv_wr); 3372 } 3373 3374 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private, 3375 int nr_cqe, int comp_vector, 3376 enum ib_poll_context poll_ctx, const char *caller); 3377 #define ib_alloc_cq(device, priv, nr_cqe, comp_vect, poll_ctx) \ 3378 __ib_alloc_cq((device), (priv), (nr_cqe), (comp_vect), (poll_ctx), KBUILD_MODNAME) 3379 3380 void ib_free_cq(struct ib_cq *cq); 3381 int ib_process_cq_direct(struct ib_cq *cq, int budget); 3382 3383 /** 3384 * ib_create_cq - Creates a CQ on the specified device. 3385 * @device: The device on which to create the CQ. 3386 * @comp_handler: A user-specified callback that is invoked when a 3387 * completion event occurs on the CQ. 3388 * @event_handler: A user-specified callback that is invoked when an 3389 * asynchronous event not associated with a completion occurs on the CQ. 3390 * @cq_context: Context associated with the CQ returned to the user via 3391 * the associated completion and event handlers. 3392 * @cq_attr: The attributes the CQ should be created upon. 3393 * 3394 * Users can examine the cq structure to determine the actual CQ size. 3395 */ 3396 struct ib_cq *ib_create_cq(struct ib_device *device, 3397 ib_comp_handler comp_handler, 3398 void (*event_handler)(struct ib_event *, void *), 3399 void *cq_context, 3400 const struct ib_cq_init_attr *cq_attr); 3401 3402 /** 3403 * ib_resize_cq - Modifies the capacity of the CQ. 3404 * @cq: The CQ to resize. 3405 * @cqe: The minimum size of the CQ. 3406 * 3407 * Users can examine the cq structure to determine the actual CQ size. 3408 */ 3409 int ib_resize_cq(struct ib_cq *cq, int cqe); 3410 3411 /** 3412 * rdma_set_cq_moderation - Modifies moderation params of the CQ 3413 * @cq: The CQ to modify. 3414 * @cq_count: number of CQEs that will trigger an event 3415 * @cq_period: max period of time in usec before triggering an event 3416 * 3417 */ 3418 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period); 3419 3420 /** 3421 * ib_destroy_cq - Destroys the specified CQ. 3422 * @cq: The CQ to destroy. 3423 */ 3424 int ib_destroy_cq(struct ib_cq *cq); 3425 3426 /** 3427 * ib_poll_cq - poll a CQ for completion(s) 3428 * @cq:the CQ being polled 3429 * @num_entries:maximum number of completions to return 3430 * @wc:array of at least @num_entries &struct ib_wc where completions 3431 * will be returned 3432 * 3433 * Poll a CQ for (possibly multiple) completions. If the return value 3434 * is < 0, an error occurred. If the return value is >= 0, it is the 3435 * number of completions returned. If the return value is 3436 * non-negative and < num_entries, then the CQ was emptied. 3437 */ 3438 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries, 3439 struct ib_wc *wc) 3440 { 3441 return cq->device->poll_cq(cq, num_entries, wc); 3442 } 3443 3444 /** 3445 * ib_req_notify_cq - Request completion notification on a CQ. 3446 * @cq: The CQ to generate an event for. 3447 * @flags: 3448 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP 3449 * to request an event on the next solicited event or next work 3450 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS 3451 * may also be |ed in to request a hint about missed events, as 3452 * described below. 3453 * 3454 * Return Value: 3455 * < 0 means an error occurred while requesting notification 3456 * == 0 means notification was requested successfully, and if 3457 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events 3458 * were missed and it is safe to wait for another event. In 3459 * this case is it guaranteed that any work completions added 3460 * to the CQ since the last CQ poll will trigger a completion 3461 * notification event. 3462 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed 3463 * in. It means that the consumer must poll the CQ again to 3464 * make sure it is empty to avoid missing an event because of a 3465 * race between requesting notification and an entry being 3466 * added to the CQ. This return value means it is possible 3467 * (but not guaranteed) that a work completion has been added 3468 * to the CQ since the last poll without triggering a 3469 * completion notification event. 3470 */ 3471 static inline int ib_req_notify_cq(struct ib_cq *cq, 3472 enum ib_cq_notify_flags flags) 3473 { 3474 return cq->device->req_notify_cq(cq, flags); 3475 } 3476 3477 /** 3478 * ib_req_ncomp_notif - Request completion notification when there are 3479 * at least the specified number of unreaped completions on the CQ. 3480 * @cq: The CQ to generate an event for. 3481 * @wc_cnt: The number of unreaped completions that should be on the 3482 * CQ before an event is generated. 3483 */ 3484 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt) 3485 { 3486 return cq->device->req_ncomp_notif ? 3487 cq->device->req_ncomp_notif(cq, wc_cnt) : 3488 -ENOSYS; 3489 } 3490 3491 /** 3492 * ib_dma_mapping_error - check a DMA addr for error 3493 * @dev: The device for which the dma_addr was created 3494 * @dma_addr: The DMA address to check 3495 */ 3496 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr) 3497 { 3498 return dma_mapping_error(dev->dma_device, dma_addr); 3499 } 3500 3501 /** 3502 * ib_dma_map_single - Map a kernel virtual address to DMA address 3503 * @dev: The device for which the dma_addr is to be created 3504 * @cpu_addr: The kernel virtual address 3505 * @size: The size of the region in bytes 3506 * @direction: The direction of the DMA 3507 */ 3508 static inline u64 ib_dma_map_single(struct ib_device *dev, 3509 void *cpu_addr, size_t size, 3510 enum dma_data_direction direction) 3511 { 3512 return dma_map_single(dev->dma_device, cpu_addr, size, direction); 3513 } 3514 3515 /** 3516 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single() 3517 * @dev: The device for which the DMA address was created 3518 * @addr: The DMA address 3519 * @size: The size of the region in bytes 3520 * @direction: The direction of the DMA 3521 */ 3522 static inline void ib_dma_unmap_single(struct ib_device *dev, 3523 u64 addr, size_t size, 3524 enum dma_data_direction direction) 3525 { 3526 dma_unmap_single(dev->dma_device, addr, size, direction); 3527 } 3528 3529 /** 3530 * ib_dma_map_page - Map a physical page to DMA address 3531 * @dev: The device for which the dma_addr is to be created 3532 * @page: The page to be mapped 3533 * @offset: The offset within the page 3534 * @size: The size of the region in bytes 3535 * @direction: The direction of the DMA 3536 */ 3537 static inline u64 ib_dma_map_page(struct ib_device *dev, 3538 struct page *page, 3539 unsigned long offset, 3540 size_t size, 3541 enum dma_data_direction direction) 3542 { 3543 return dma_map_page(dev->dma_device, page, offset, size, direction); 3544 } 3545 3546 /** 3547 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page() 3548 * @dev: The device for which the DMA address was created 3549 * @addr: The DMA address 3550 * @size: The size of the region in bytes 3551 * @direction: The direction of the DMA 3552 */ 3553 static inline void ib_dma_unmap_page(struct ib_device *dev, 3554 u64 addr, size_t size, 3555 enum dma_data_direction direction) 3556 { 3557 dma_unmap_page(dev->dma_device, addr, size, direction); 3558 } 3559 3560 /** 3561 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses 3562 * @dev: The device for which the DMA addresses are to be created 3563 * @sg: The array of scatter/gather entries 3564 * @nents: The number of scatter/gather entries 3565 * @direction: The direction of the DMA 3566 */ 3567 static inline int ib_dma_map_sg(struct ib_device *dev, 3568 struct scatterlist *sg, int nents, 3569 enum dma_data_direction direction) 3570 { 3571 return dma_map_sg(dev->dma_device, sg, nents, direction); 3572 } 3573 3574 /** 3575 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses 3576 * @dev: The device for which the DMA addresses were created 3577 * @sg: The array of scatter/gather entries 3578 * @nents: The number of scatter/gather entries 3579 * @direction: The direction of the DMA 3580 */ 3581 static inline void ib_dma_unmap_sg(struct ib_device *dev, 3582 struct scatterlist *sg, int nents, 3583 enum dma_data_direction direction) 3584 { 3585 dma_unmap_sg(dev->dma_device, sg, nents, direction); 3586 } 3587 3588 static inline int ib_dma_map_sg_attrs(struct ib_device *dev, 3589 struct scatterlist *sg, int nents, 3590 enum dma_data_direction direction, 3591 unsigned long dma_attrs) 3592 { 3593 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, 3594 dma_attrs); 3595 } 3596 3597 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev, 3598 struct scatterlist *sg, int nents, 3599 enum dma_data_direction direction, 3600 unsigned long dma_attrs) 3601 { 3602 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs); 3603 } 3604 /** 3605 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry 3606 * @dev: The device for which the DMA addresses were created 3607 * @sg: The scatter/gather entry 3608 * 3609 * Note: this function is obsolete. To do: change all occurrences of 3610 * ib_sg_dma_address() into sg_dma_address(). 3611 */ 3612 static inline u64 ib_sg_dma_address(struct ib_device *dev, 3613 struct scatterlist *sg) 3614 { 3615 return sg_dma_address(sg); 3616 } 3617 3618 /** 3619 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry 3620 * @dev: The device for which the DMA addresses were created 3621 * @sg: The scatter/gather entry 3622 * 3623 * Note: this function is obsolete. To do: change all occurrences of 3624 * ib_sg_dma_len() into sg_dma_len(). 3625 */ 3626 static inline unsigned int ib_sg_dma_len(struct ib_device *dev, 3627 struct scatterlist *sg) 3628 { 3629 return sg_dma_len(sg); 3630 } 3631 3632 /** 3633 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU 3634 * @dev: The device for which the DMA address was created 3635 * @addr: The DMA address 3636 * @size: The size of the region in bytes 3637 * @dir: The direction of the DMA 3638 */ 3639 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev, 3640 u64 addr, 3641 size_t size, 3642 enum dma_data_direction dir) 3643 { 3644 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir); 3645 } 3646 3647 /** 3648 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device 3649 * @dev: The device for which the DMA address was created 3650 * @addr: The DMA address 3651 * @size: The size of the region in bytes 3652 * @dir: The direction of the DMA 3653 */ 3654 static inline void ib_dma_sync_single_for_device(struct ib_device *dev, 3655 u64 addr, 3656 size_t size, 3657 enum dma_data_direction dir) 3658 { 3659 dma_sync_single_for_device(dev->dma_device, addr, size, dir); 3660 } 3661 3662 /** 3663 * ib_dma_alloc_coherent - Allocate memory and map it for DMA 3664 * @dev: The device for which the DMA address is requested 3665 * @size: The size of the region to allocate in bytes 3666 * @dma_handle: A pointer for returning the DMA address of the region 3667 * @flag: memory allocator flags 3668 */ 3669 static inline void *ib_dma_alloc_coherent(struct ib_device *dev, 3670 size_t size, 3671 dma_addr_t *dma_handle, 3672 gfp_t flag) 3673 { 3674 return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag); 3675 } 3676 3677 /** 3678 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent() 3679 * @dev: The device for which the DMA addresses were allocated 3680 * @size: The size of the region 3681 * @cpu_addr: the address returned by ib_dma_alloc_coherent() 3682 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent() 3683 */ 3684 static inline void ib_dma_free_coherent(struct ib_device *dev, 3685 size_t size, void *cpu_addr, 3686 dma_addr_t dma_handle) 3687 { 3688 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle); 3689 } 3690 3691 /** 3692 * ib_dereg_mr - Deregisters a memory region and removes it from the 3693 * HCA translation table. 3694 * @mr: The memory region to deregister. 3695 * 3696 * This function can fail, if the memory region has memory windows bound to it. 3697 */ 3698 int ib_dereg_mr(struct ib_mr *mr); 3699 3700 struct ib_mr *ib_alloc_mr(struct ib_pd *pd, 3701 enum ib_mr_type mr_type, 3702 u32 max_num_sg); 3703 3704 /** 3705 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR 3706 * R_Key and L_Key. 3707 * @mr - struct ib_mr pointer to be updated. 3708 * @newkey - new key to be used. 3709 */ 3710 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey) 3711 { 3712 mr->lkey = (mr->lkey & 0xffffff00) | newkey; 3713 mr->rkey = (mr->rkey & 0xffffff00) | newkey; 3714 } 3715 3716 /** 3717 * ib_inc_rkey - increments the key portion of the given rkey. Can be used 3718 * for calculating a new rkey for type 2 memory windows. 3719 * @rkey - the rkey to increment. 3720 */ 3721 static inline u32 ib_inc_rkey(u32 rkey) 3722 { 3723 const u32 mask = 0x000000ff; 3724 return ((rkey + 1) & mask) | (rkey & ~mask); 3725 } 3726 3727 /** 3728 * ib_alloc_fmr - Allocates a unmapped fast memory region. 3729 * @pd: The protection domain associated with the unmapped region. 3730 * @mr_access_flags: Specifies the memory access rights. 3731 * @fmr_attr: Attributes of the unmapped region. 3732 * 3733 * A fast memory region must be mapped before it can be used as part of 3734 * a work request. 3735 */ 3736 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd, 3737 int mr_access_flags, 3738 struct ib_fmr_attr *fmr_attr); 3739 3740 /** 3741 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region. 3742 * @fmr: The fast memory region to associate with the pages. 3743 * @page_list: An array of physical pages to map to the fast memory region. 3744 * @list_len: The number of pages in page_list. 3745 * @iova: The I/O virtual address to use with the mapped region. 3746 */ 3747 static inline int ib_map_phys_fmr(struct ib_fmr *fmr, 3748 u64 *page_list, int list_len, 3749 u64 iova) 3750 { 3751 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova); 3752 } 3753 3754 /** 3755 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions. 3756 * @fmr_list: A linked list of fast memory regions to unmap. 3757 */ 3758 int ib_unmap_fmr(struct list_head *fmr_list); 3759 3760 /** 3761 * ib_dealloc_fmr - Deallocates a fast memory region. 3762 * @fmr: The fast memory region to deallocate. 3763 */ 3764 int ib_dealloc_fmr(struct ib_fmr *fmr); 3765 3766 /** 3767 * ib_attach_mcast - Attaches the specified QP to a multicast group. 3768 * @qp: QP to attach to the multicast group. The QP must be type 3769 * IB_QPT_UD. 3770 * @gid: Multicast group GID. 3771 * @lid: Multicast group LID in host byte order. 3772 * 3773 * In order to send and receive multicast packets, subnet 3774 * administration must have created the multicast group and configured 3775 * the fabric appropriately. The port associated with the specified 3776 * QP must also be a member of the multicast group. 3777 */ 3778 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid); 3779 3780 /** 3781 * ib_detach_mcast - Detaches the specified QP from a multicast group. 3782 * @qp: QP to detach from the multicast group. 3783 * @gid: Multicast group GID. 3784 * @lid: Multicast group LID in host byte order. 3785 */ 3786 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid); 3787 3788 /** 3789 * ib_alloc_xrcd - Allocates an XRC domain. 3790 * @device: The device on which to allocate the XRC domain. 3791 * @caller: Module name for kernel consumers 3792 */ 3793 struct ib_xrcd *__ib_alloc_xrcd(struct ib_device *device, const char *caller); 3794 #define ib_alloc_xrcd(device) \ 3795 __ib_alloc_xrcd((device), KBUILD_MODNAME) 3796 3797 /** 3798 * ib_dealloc_xrcd - Deallocates an XRC domain. 3799 * @xrcd: The XRC domain to deallocate. 3800 */ 3801 int ib_dealloc_xrcd(struct ib_xrcd *xrcd); 3802 3803 struct ib_flow *ib_create_flow(struct ib_qp *qp, 3804 struct ib_flow_attr *flow_attr, int domain); 3805 int ib_destroy_flow(struct ib_flow *flow_id); 3806 3807 static inline int ib_check_mr_access(int flags) 3808 { 3809 /* 3810 * Local write permission is required if remote write or 3811 * remote atomic permission is also requested. 3812 */ 3813 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) && 3814 !(flags & IB_ACCESS_LOCAL_WRITE)) 3815 return -EINVAL; 3816 3817 return 0; 3818 } 3819 3820 static inline bool ib_access_writable(int access_flags) 3821 { 3822 /* 3823 * We have writable memory backing the MR if any of the following 3824 * access flags are set. "Local write" and "remote write" obviously 3825 * require write access. "Remote atomic" can do things like fetch and 3826 * add, which will modify memory, and "MW bind" can change permissions 3827 * by binding a window. 3828 */ 3829 return access_flags & 3830 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE | 3831 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND); 3832 } 3833 3834 /** 3835 * ib_check_mr_status: lightweight check of MR status. 3836 * This routine may provide status checks on a selected 3837 * ib_mr. first use is for signature status check. 3838 * 3839 * @mr: A memory region. 3840 * @check_mask: Bitmask of which checks to perform from 3841 * ib_mr_status_check enumeration. 3842 * @mr_status: The container of relevant status checks. 3843 * failed checks will be indicated in the status bitmask 3844 * and the relevant info shall be in the error item. 3845 */ 3846 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask, 3847 struct ib_mr_status *mr_status); 3848 3849 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port, 3850 u16 pkey, const union ib_gid *gid, 3851 const struct sockaddr *addr); 3852 struct ib_wq *ib_create_wq(struct ib_pd *pd, 3853 struct ib_wq_init_attr *init_attr); 3854 int ib_destroy_wq(struct ib_wq *wq); 3855 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr, 3856 u32 wq_attr_mask); 3857 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device, 3858 struct ib_rwq_ind_table_init_attr* 3859 wq_ind_table_init_attr); 3860 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table); 3861 3862 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents, 3863 unsigned int *sg_offset, unsigned int page_size); 3864 3865 static inline int 3866 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents, 3867 unsigned int *sg_offset, unsigned int page_size) 3868 { 3869 int n; 3870 3871 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size); 3872 mr->iova = 0; 3873 3874 return n; 3875 } 3876 3877 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents, 3878 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64)); 3879 3880 void ib_drain_rq(struct ib_qp *qp); 3881 void ib_drain_sq(struct ib_qp *qp); 3882 void ib_drain_qp(struct ib_qp *qp); 3883 3884 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width); 3885 3886 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr) 3887 { 3888 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE) 3889 return attr->roce.dmac; 3890 return NULL; 3891 } 3892 3893 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid) 3894 { 3895 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 3896 attr->ib.dlid = (u16)dlid; 3897 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 3898 attr->opa.dlid = dlid; 3899 } 3900 3901 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr) 3902 { 3903 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 3904 return attr->ib.dlid; 3905 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 3906 return attr->opa.dlid; 3907 return 0; 3908 } 3909 3910 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl) 3911 { 3912 attr->sl = sl; 3913 } 3914 3915 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr) 3916 { 3917 return attr->sl; 3918 } 3919 3920 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr, 3921 u8 src_path_bits) 3922 { 3923 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 3924 attr->ib.src_path_bits = src_path_bits; 3925 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 3926 attr->opa.src_path_bits = src_path_bits; 3927 } 3928 3929 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr) 3930 { 3931 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 3932 return attr->ib.src_path_bits; 3933 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 3934 return attr->opa.src_path_bits; 3935 return 0; 3936 } 3937 3938 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr, 3939 bool make_grd) 3940 { 3941 if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 3942 attr->opa.make_grd = make_grd; 3943 } 3944 3945 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr) 3946 { 3947 if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 3948 return attr->opa.make_grd; 3949 return false; 3950 } 3951 3952 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num) 3953 { 3954 attr->port_num = port_num; 3955 } 3956 3957 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr) 3958 { 3959 return attr->port_num; 3960 } 3961 3962 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr, 3963 u8 static_rate) 3964 { 3965 attr->static_rate = static_rate; 3966 } 3967 3968 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr) 3969 { 3970 return attr->static_rate; 3971 } 3972 3973 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr, 3974 enum ib_ah_flags flag) 3975 { 3976 attr->ah_flags = flag; 3977 } 3978 3979 static inline enum ib_ah_flags 3980 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr) 3981 { 3982 return attr->ah_flags; 3983 } 3984 3985 static inline const struct ib_global_route 3986 *rdma_ah_read_grh(const struct rdma_ah_attr *attr) 3987 { 3988 return &attr->grh; 3989 } 3990 3991 /*To retrieve and modify the grh */ 3992 static inline struct ib_global_route 3993 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr) 3994 { 3995 return &attr->grh; 3996 } 3997 3998 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid) 3999 { 4000 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 4001 4002 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid)); 4003 } 4004 4005 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr, 4006 __be64 prefix) 4007 { 4008 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 4009 4010 grh->dgid.global.subnet_prefix = prefix; 4011 } 4012 4013 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr, 4014 __be64 if_id) 4015 { 4016 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 4017 4018 grh->dgid.global.interface_id = if_id; 4019 } 4020 4021 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr, 4022 union ib_gid *dgid, u32 flow_label, 4023 u8 sgid_index, u8 hop_limit, 4024 u8 traffic_class) 4025 { 4026 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 4027 4028 attr->ah_flags = IB_AH_GRH; 4029 if (dgid) 4030 grh->dgid = *dgid; 4031 grh->flow_label = flow_label; 4032 grh->sgid_index = sgid_index; 4033 grh->hop_limit = hop_limit; 4034 grh->traffic_class = traffic_class; 4035 grh->sgid_attr = NULL; 4036 } 4037 4038 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr); 4039 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid, 4040 u32 flow_label, u8 hop_limit, u8 traffic_class, 4041 const struct ib_gid_attr *sgid_attr); 4042 void rdma_copy_ah_attr(struct rdma_ah_attr *dest, 4043 const struct rdma_ah_attr *src); 4044 void rdma_replace_ah_attr(struct rdma_ah_attr *old, 4045 const struct rdma_ah_attr *new); 4046 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src); 4047 4048 /** 4049 * rdma_ah_find_type - Return address handle type. 4050 * 4051 * @dev: Device to be checked 4052 * @port_num: Port number 4053 */ 4054 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev, 4055 u8 port_num) 4056 { 4057 if (rdma_protocol_roce(dev, port_num)) 4058 return RDMA_AH_ATTR_TYPE_ROCE; 4059 if (rdma_protocol_ib(dev, port_num)) { 4060 if (rdma_cap_opa_ah(dev, port_num)) 4061 return RDMA_AH_ATTR_TYPE_OPA; 4062 return RDMA_AH_ATTR_TYPE_IB; 4063 } 4064 4065 return RDMA_AH_ATTR_TYPE_UNDEFINED; 4066 } 4067 4068 /** 4069 * ib_lid_cpu16 - Return lid in 16bit CPU encoding. 4070 * In the current implementation the only way to get 4071 * get the 32bit lid is from other sources for OPA. 4072 * For IB, lids will always be 16bits so cast the 4073 * value accordingly. 4074 * 4075 * @lid: A 32bit LID 4076 */ 4077 static inline u16 ib_lid_cpu16(u32 lid) 4078 { 4079 WARN_ON_ONCE(lid & 0xFFFF0000); 4080 return (u16)lid; 4081 } 4082 4083 /** 4084 * ib_lid_be16 - Return lid in 16bit BE encoding. 4085 * 4086 * @lid: A 32bit LID 4087 */ 4088 static inline __be16 ib_lid_be16(u32 lid) 4089 { 4090 WARN_ON_ONCE(lid & 0xFFFF0000); 4091 return cpu_to_be16((u16)lid); 4092 } 4093 4094 /** 4095 * ib_get_vector_affinity - Get the affinity mappings of a given completion 4096 * vector 4097 * @device: the rdma device 4098 * @comp_vector: index of completion vector 4099 * 4100 * Returns NULL on failure, otherwise a corresponding cpu map of the 4101 * completion vector (returns all-cpus map if the device driver doesn't 4102 * implement get_vector_affinity). 4103 */ 4104 static inline const struct cpumask * 4105 ib_get_vector_affinity(struct ib_device *device, int comp_vector) 4106 { 4107 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors || 4108 !device->get_vector_affinity) 4109 return NULL; 4110 4111 return device->get_vector_affinity(device, comp_vector); 4112 4113 } 4114 4115 /** 4116 * rdma_roce_rescan_device - Rescan all of the network devices in the system 4117 * and add their gids, as needed, to the relevant RoCE devices. 4118 * 4119 * @device: the rdma device 4120 */ 4121 void rdma_roce_rescan_device(struct ib_device *ibdev); 4122 4123 #endif /* IB_VERBS_H */ 4124