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