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/dma-mapping.h> 45 #include <linux/kref.h> 46 #include <linux/list.h> 47 #include <linux/rwsem.h> 48 #include <linux/workqueue.h> 49 #include <linux/irq_poll.h> 50 #include <uapi/linux/if_ether.h> 51 #include <net/ipv6.h> 52 #include <net/ip.h> 53 #include <linux/string.h> 54 #include <linux/slab.h> 55 #include <linux/netdevice.h> 56 #include <linux/refcount.h> 57 #include <linux/if_link.h> 58 #include <linux/atomic.h> 59 #include <linux/mmu_notifier.h> 60 #include <linux/uaccess.h> 61 #include <linux/cgroup_rdma.h> 62 #include <uapi/rdma/ib_user_verbs.h> 63 #include <rdma/restrack.h> 64 #include <uapi/rdma/rdma_user_ioctl.h> 65 #include <uapi/rdma/ib_user_ioctl_verbs.h> 66 67 #define IB_FW_VERSION_NAME_MAX ETHTOOL_FWVERS_LEN 68 69 struct ib_umem_odp; 70 71 extern struct workqueue_struct *ib_wq; 72 extern struct workqueue_struct *ib_comp_wq; 73 extern struct workqueue_struct *ib_comp_unbound_wq; 74 75 union ib_gid { 76 u8 raw[16]; 77 struct { 78 __be64 subnet_prefix; 79 __be64 interface_id; 80 } global; 81 }; 82 83 extern union ib_gid zgid; 84 85 enum ib_gid_type { 86 /* If link layer is Ethernet, this is RoCE V1 */ 87 IB_GID_TYPE_IB = 0, 88 IB_GID_TYPE_ROCE = 0, 89 IB_GID_TYPE_ROCE_UDP_ENCAP = 1, 90 IB_GID_TYPE_SIZE 91 }; 92 93 #define ROCE_V2_UDP_DPORT 4791 94 struct ib_gid_attr { 95 struct net_device *ndev; 96 struct ib_device *device; 97 union ib_gid gid; 98 enum ib_gid_type gid_type; 99 u16 index; 100 u8 port_num; 101 }; 102 103 enum rdma_node_type { 104 /* IB values map to NodeInfo:NodeType. */ 105 RDMA_NODE_IB_CA = 1, 106 RDMA_NODE_IB_SWITCH, 107 RDMA_NODE_IB_ROUTER, 108 RDMA_NODE_RNIC, 109 RDMA_NODE_USNIC, 110 RDMA_NODE_USNIC_UDP, 111 }; 112 113 enum { 114 /* set the local administered indication */ 115 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2, 116 }; 117 118 enum rdma_transport_type { 119 RDMA_TRANSPORT_IB, 120 RDMA_TRANSPORT_IWARP, 121 RDMA_TRANSPORT_USNIC, 122 RDMA_TRANSPORT_USNIC_UDP 123 }; 124 125 enum rdma_protocol_type { 126 RDMA_PROTOCOL_IB, 127 RDMA_PROTOCOL_IBOE, 128 RDMA_PROTOCOL_IWARP, 129 RDMA_PROTOCOL_USNIC_UDP 130 }; 131 132 __attribute_const__ enum rdma_transport_type 133 rdma_node_get_transport(enum rdma_node_type node_type); 134 135 enum rdma_network_type { 136 RDMA_NETWORK_IB, 137 RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB, 138 RDMA_NETWORK_IPV4, 139 RDMA_NETWORK_IPV6 140 }; 141 142 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type) 143 { 144 if (network_type == RDMA_NETWORK_IPV4 || 145 network_type == RDMA_NETWORK_IPV6) 146 return IB_GID_TYPE_ROCE_UDP_ENCAP; 147 148 /* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */ 149 return IB_GID_TYPE_IB; 150 } 151 152 static inline enum rdma_network_type 153 rdma_gid_attr_network_type(const struct ib_gid_attr *attr) 154 { 155 if (attr->gid_type == IB_GID_TYPE_IB) 156 return RDMA_NETWORK_IB; 157 158 if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid)) 159 return RDMA_NETWORK_IPV4; 160 else 161 return RDMA_NETWORK_IPV6; 162 } 163 164 enum rdma_link_layer { 165 IB_LINK_LAYER_UNSPECIFIED, 166 IB_LINK_LAYER_INFINIBAND, 167 IB_LINK_LAYER_ETHERNET, 168 }; 169 170 enum ib_device_cap_flags { 171 IB_DEVICE_RESIZE_MAX_WR = (1 << 0), 172 IB_DEVICE_BAD_PKEY_CNTR = (1 << 1), 173 IB_DEVICE_BAD_QKEY_CNTR = (1 << 2), 174 IB_DEVICE_RAW_MULTI = (1 << 3), 175 IB_DEVICE_AUTO_PATH_MIG = (1 << 4), 176 IB_DEVICE_CHANGE_PHY_PORT = (1 << 5), 177 IB_DEVICE_UD_AV_PORT_ENFORCE = (1 << 6), 178 IB_DEVICE_CURR_QP_STATE_MOD = (1 << 7), 179 IB_DEVICE_SHUTDOWN_PORT = (1 << 8), 180 /* Not in use, former INIT_TYPE = (1 << 9),*/ 181 IB_DEVICE_PORT_ACTIVE_EVENT = (1 << 10), 182 IB_DEVICE_SYS_IMAGE_GUID = (1 << 11), 183 IB_DEVICE_RC_RNR_NAK_GEN = (1 << 12), 184 IB_DEVICE_SRQ_RESIZE = (1 << 13), 185 IB_DEVICE_N_NOTIFY_CQ = (1 << 14), 186 187 /* 188 * This device supports a per-device lkey or stag that can be 189 * used without performing a memory registration for the local 190 * memory. Note that ULPs should never check this flag, but 191 * instead of use the local_dma_lkey flag in the ib_pd structure, 192 * which will always contain a usable lkey. 193 */ 194 IB_DEVICE_LOCAL_DMA_LKEY = (1 << 15), 195 /* Reserved, old SEND_W_INV = (1 << 16),*/ 196 IB_DEVICE_MEM_WINDOW = (1 << 17), 197 /* 198 * Devices should set IB_DEVICE_UD_IP_SUM if they support 199 * insertion of UDP and TCP checksum on outgoing UD IPoIB 200 * messages and can verify the validity of checksum for 201 * incoming messages. Setting this flag implies that the 202 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode. 203 */ 204 IB_DEVICE_UD_IP_CSUM = (1 << 18), 205 IB_DEVICE_UD_TSO = (1 << 19), 206 IB_DEVICE_XRC = (1 << 20), 207 208 /* 209 * This device supports the IB "base memory management extension", 210 * which includes support for fast registrations (IB_WR_REG_MR, 211 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should 212 * also be set by any iWarp device which must support FRs to comply 213 * to the iWarp verbs spec. iWarp devices also support the 214 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the 215 * stag. 216 */ 217 IB_DEVICE_MEM_MGT_EXTENSIONS = (1 << 21), 218 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1 << 22), 219 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1 << 23), 220 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1 << 24), 221 IB_DEVICE_RC_IP_CSUM = (1 << 25), 222 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */ 223 IB_DEVICE_RAW_IP_CSUM = (1 << 26), 224 /* 225 * Devices should set IB_DEVICE_CROSS_CHANNEL if they 226 * support execution of WQEs that involve synchronization 227 * of I/O operations with single completion queue managed 228 * by hardware. 229 */ 230 IB_DEVICE_CROSS_CHANNEL = (1 << 27), 231 IB_DEVICE_MANAGED_FLOW_STEERING = (1 << 29), 232 IB_DEVICE_SIGNATURE_HANDOVER = (1 << 30), 233 IB_DEVICE_ON_DEMAND_PAGING = (1ULL << 31), 234 IB_DEVICE_SG_GAPS_REG = (1ULL << 32), 235 IB_DEVICE_VIRTUAL_FUNCTION = (1ULL << 33), 236 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */ 237 IB_DEVICE_RAW_SCATTER_FCS = (1ULL << 34), 238 IB_DEVICE_RDMA_NETDEV_OPA_VNIC = (1ULL << 35), 239 /* The device supports padding incoming writes to cacheline. */ 240 IB_DEVICE_PCI_WRITE_END_PADDING = (1ULL << 36), 241 }; 242 243 enum ib_signature_prot_cap { 244 IB_PROT_T10DIF_TYPE_1 = 1, 245 IB_PROT_T10DIF_TYPE_2 = 1 << 1, 246 IB_PROT_T10DIF_TYPE_3 = 1 << 2, 247 }; 248 249 enum ib_signature_guard_cap { 250 IB_GUARD_T10DIF_CRC = 1, 251 IB_GUARD_T10DIF_CSUM = 1 << 1, 252 }; 253 254 enum ib_atomic_cap { 255 IB_ATOMIC_NONE, 256 IB_ATOMIC_HCA, 257 IB_ATOMIC_GLOB 258 }; 259 260 enum ib_odp_general_cap_bits { 261 IB_ODP_SUPPORT = 1 << 0, 262 IB_ODP_SUPPORT_IMPLICIT = 1 << 1, 263 }; 264 265 enum ib_odp_transport_cap_bits { 266 IB_ODP_SUPPORT_SEND = 1 << 0, 267 IB_ODP_SUPPORT_RECV = 1 << 1, 268 IB_ODP_SUPPORT_WRITE = 1 << 2, 269 IB_ODP_SUPPORT_READ = 1 << 3, 270 IB_ODP_SUPPORT_ATOMIC = 1 << 4, 271 }; 272 273 struct ib_odp_caps { 274 uint64_t general_caps; 275 struct { 276 uint32_t rc_odp_caps; 277 uint32_t uc_odp_caps; 278 uint32_t ud_odp_caps; 279 } per_transport_caps; 280 }; 281 282 struct ib_rss_caps { 283 /* Corresponding bit will be set if qp type from 284 * 'enum ib_qp_type' is supported, e.g. 285 * supported_qpts |= 1 << IB_QPT_UD 286 */ 287 u32 supported_qpts; 288 u32 max_rwq_indirection_tables; 289 u32 max_rwq_indirection_table_size; 290 }; 291 292 enum ib_tm_cap_flags { 293 /* Support tag matching on RC transport */ 294 IB_TM_CAP_RC = 1 << 0, 295 }; 296 297 struct ib_tm_caps { 298 /* Max size of RNDV header */ 299 u32 max_rndv_hdr_size; 300 /* Max number of entries in tag matching list */ 301 u32 max_num_tags; 302 /* From enum ib_tm_cap_flags */ 303 u32 flags; 304 /* Max number of outstanding list operations */ 305 u32 max_ops; 306 /* Max number of SGE in tag matching entry */ 307 u32 max_sge; 308 }; 309 310 struct ib_cq_init_attr { 311 unsigned int cqe; 312 int comp_vector; 313 u32 flags; 314 }; 315 316 enum ib_cq_attr_mask { 317 IB_CQ_MODERATE = 1 << 0, 318 }; 319 320 struct ib_cq_caps { 321 u16 max_cq_moderation_count; 322 u16 max_cq_moderation_period; 323 }; 324 325 struct ib_dm_mr_attr { 326 u64 length; 327 u64 offset; 328 u32 access_flags; 329 }; 330 331 struct ib_dm_alloc_attr { 332 u64 length; 333 u32 alignment; 334 u32 flags; 335 }; 336 337 struct ib_device_attr { 338 u64 fw_ver; 339 __be64 sys_image_guid; 340 u64 max_mr_size; 341 u64 page_size_cap; 342 u32 vendor_id; 343 u32 vendor_part_id; 344 u32 hw_ver; 345 int max_qp; 346 int max_qp_wr; 347 u64 device_cap_flags; 348 int max_send_sge; 349 int max_recv_sge; 350 int max_sge_rd; 351 int max_cq; 352 int max_cqe; 353 int max_mr; 354 int max_pd; 355 int max_qp_rd_atom; 356 int max_ee_rd_atom; 357 int max_res_rd_atom; 358 int max_qp_init_rd_atom; 359 int max_ee_init_rd_atom; 360 enum ib_atomic_cap atomic_cap; 361 enum ib_atomic_cap masked_atomic_cap; 362 int max_ee; 363 int max_rdd; 364 int max_mw; 365 int max_raw_ipv6_qp; 366 int max_raw_ethy_qp; 367 int max_mcast_grp; 368 int max_mcast_qp_attach; 369 int max_total_mcast_qp_attach; 370 int max_ah; 371 int max_fmr; 372 int max_map_per_fmr; 373 int max_srq; 374 int max_srq_wr; 375 int max_srq_sge; 376 unsigned int max_fast_reg_page_list_len; 377 u16 max_pkeys; 378 u8 local_ca_ack_delay; 379 int sig_prot_cap; 380 int sig_guard_cap; 381 struct ib_odp_caps odp_caps; 382 uint64_t timestamp_mask; 383 uint64_t hca_core_clock; /* in KHZ */ 384 struct ib_rss_caps rss_caps; 385 u32 max_wq_type_rq; 386 u32 raw_packet_caps; /* Use ib_raw_packet_caps enum */ 387 struct ib_tm_caps tm_caps; 388 struct ib_cq_caps cq_caps; 389 u64 max_dm_size; 390 }; 391 392 enum ib_mtu { 393 IB_MTU_256 = 1, 394 IB_MTU_512 = 2, 395 IB_MTU_1024 = 3, 396 IB_MTU_2048 = 4, 397 IB_MTU_4096 = 5 398 }; 399 400 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu) 401 { 402 switch (mtu) { 403 case IB_MTU_256: return 256; 404 case IB_MTU_512: return 512; 405 case IB_MTU_1024: return 1024; 406 case IB_MTU_2048: return 2048; 407 case IB_MTU_4096: return 4096; 408 default: return -1; 409 } 410 } 411 412 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu) 413 { 414 if (mtu >= 4096) 415 return IB_MTU_4096; 416 else if (mtu >= 2048) 417 return IB_MTU_2048; 418 else if (mtu >= 1024) 419 return IB_MTU_1024; 420 else if (mtu >= 512) 421 return IB_MTU_512; 422 else 423 return IB_MTU_256; 424 } 425 426 enum ib_port_state { 427 IB_PORT_NOP = 0, 428 IB_PORT_DOWN = 1, 429 IB_PORT_INIT = 2, 430 IB_PORT_ARMED = 3, 431 IB_PORT_ACTIVE = 4, 432 IB_PORT_ACTIVE_DEFER = 5 433 }; 434 435 enum ib_port_width { 436 IB_WIDTH_1X = 1, 437 IB_WIDTH_2X = 16, 438 IB_WIDTH_4X = 2, 439 IB_WIDTH_8X = 4, 440 IB_WIDTH_12X = 8 441 }; 442 443 static inline int ib_width_enum_to_int(enum ib_port_width width) 444 { 445 switch (width) { 446 case IB_WIDTH_1X: return 1; 447 case IB_WIDTH_2X: return 2; 448 case IB_WIDTH_4X: return 4; 449 case IB_WIDTH_8X: return 8; 450 case IB_WIDTH_12X: return 12; 451 default: return -1; 452 } 453 } 454 455 enum ib_port_speed { 456 IB_SPEED_SDR = 1, 457 IB_SPEED_DDR = 2, 458 IB_SPEED_QDR = 4, 459 IB_SPEED_FDR10 = 8, 460 IB_SPEED_FDR = 16, 461 IB_SPEED_EDR = 32, 462 IB_SPEED_HDR = 64 463 }; 464 465 /** 466 * struct rdma_hw_stats 467 * @lock - Mutex to protect parallel write access to lifespan and values 468 * of counters, which are 64bits and not guaranteeed to be written 469 * atomicaly on 32bits systems. 470 * @timestamp - Used by the core code to track when the last update was 471 * @lifespan - Used by the core code to determine how old the counters 472 * should be before being updated again. Stored in jiffies, defaults 473 * to 10 milliseconds, drivers can override the default be specifying 474 * their own value during their allocation routine. 475 * @name - Array of pointers to static names used for the counters in 476 * directory. 477 * @num_counters - How many hardware counters there are. If name is 478 * shorter than this number, a kernel oops will result. Driver authors 479 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters) 480 * in their code to prevent this. 481 * @value - Array of u64 counters that are accessed by the sysfs code and 482 * filled in by the drivers get_stats routine 483 */ 484 struct rdma_hw_stats { 485 struct mutex lock; /* Protect lifespan and values[] */ 486 unsigned long timestamp; 487 unsigned long lifespan; 488 const char * const *names; 489 int num_counters; 490 u64 value[]; 491 }; 492 493 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10 494 /** 495 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct 496 * for drivers. 497 * @names - Array of static const char * 498 * @num_counters - How many elements in array 499 * @lifespan - How many milliseconds between updates 500 */ 501 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct( 502 const char * const *names, int num_counters, 503 unsigned long lifespan) 504 { 505 struct rdma_hw_stats *stats; 506 507 stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64), 508 GFP_KERNEL); 509 if (!stats) 510 return NULL; 511 stats->names = names; 512 stats->num_counters = num_counters; 513 stats->lifespan = msecs_to_jiffies(lifespan); 514 515 return stats; 516 } 517 518 519 /* Define bits for the various functionality this port needs to be supported by 520 * the core. 521 */ 522 /* Management 0x00000FFF */ 523 #define RDMA_CORE_CAP_IB_MAD 0x00000001 524 #define RDMA_CORE_CAP_IB_SMI 0x00000002 525 #define RDMA_CORE_CAP_IB_CM 0x00000004 526 #define RDMA_CORE_CAP_IW_CM 0x00000008 527 #define RDMA_CORE_CAP_IB_SA 0x00000010 528 #define RDMA_CORE_CAP_OPA_MAD 0x00000020 529 530 /* Address format 0x000FF000 */ 531 #define RDMA_CORE_CAP_AF_IB 0x00001000 532 #define RDMA_CORE_CAP_ETH_AH 0x00002000 533 #define RDMA_CORE_CAP_OPA_AH 0x00004000 534 #define RDMA_CORE_CAP_IB_GRH_REQUIRED 0x00008000 535 536 /* Protocol 0xFFF00000 */ 537 #define RDMA_CORE_CAP_PROT_IB 0x00100000 538 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000 539 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000 540 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000 541 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000 542 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000 543 544 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \ 545 | RDMA_CORE_CAP_PROT_ROCE \ 546 | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP) 547 548 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \ 549 | RDMA_CORE_CAP_IB_MAD \ 550 | RDMA_CORE_CAP_IB_SMI \ 551 | RDMA_CORE_CAP_IB_CM \ 552 | RDMA_CORE_CAP_IB_SA \ 553 | RDMA_CORE_CAP_AF_IB) 554 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \ 555 | RDMA_CORE_CAP_IB_MAD \ 556 | RDMA_CORE_CAP_IB_CM \ 557 | RDMA_CORE_CAP_AF_IB \ 558 | RDMA_CORE_CAP_ETH_AH) 559 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \ 560 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \ 561 | RDMA_CORE_CAP_IB_MAD \ 562 | RDMA_CORE_CAP_IB_CM \ 563 | RDMA_CORE_CAP_AF_IB \ 564 | RDMA_CORE_CAP_ETH_AH) 565 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \ 566 | RDMA_CORE_CAP_IW_CM) 567 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \ 568 | RDMA_CORE_CAP_OPA_MAD) 569 570 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET) 571 572 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC) 573 574 struct ib_port_attr { 575 u64 subnet_prefix; 576 enum ib_port_state state; 577 enum ib_mtu max_mtu; 578 enum ib_mtu active_mtu; 579 int gid_tbl_len; 580 unsigned int ip_gids:1; 581 /* This is the value from PortInfo CapabilityMask, defined by IBA */ 582 u32 port_cap_flags; 583 u32 max_msg_sz; 584 u32 bad_pkey_cntr; 585 u32 qkey_viol_cntr; 586 u16 pkey_tbl_len; 587 u32 sm_lid; 588 u32 lid; 589 u8 lmc; 590 u8 max_vl_num; 591 u8 sm_sl; 592 u8 subnet_timeout; 593 u8 init_type_reply; 594 u8 active_width; 595 u8 active_speed; 596 u8 phys_state; 597 u16 port_cap_flags2; 598 }; 599 600 enum ib_device_modify_flags { 601 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0, 602 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1 603 }; 604 605 #define IB_DEVICE_NODE_DESC_MAX 64 606 607 struct ib_device_modify { 608 u64 sys_image_guid; 609 char node_desc[IB_DEVICE_NODE_DESC_MAX]; 610 }; 611 612 enum ib_port_modify_flags { 613 IB_PORT_SHUTDOWN = 1, 614 IB_PORT_INIT_TYPE = (1<<2), 615 IB_PORT_RESET_QKEY_CNTR = (1<<3), 616 IB_PORT_OPA_MASK_CHG = (1<<4) 617 }; 618 619 struct ib_port_modify { 620 u32 set_port_cap_mask; 621 u32 clr_port_cap_mask; 622 u8 init_type; 623 }; 624 625 enum ib_event_type { 626 IB_EVENT_CQ_ERR, 627 IB_EVENT_QP_FATAL, 628 IB_EVENT_QP_REQ_ERR, 629 IB_EVENT_QP_ACCESS_ERR, 630 IB_EVENT_COMM_EST, 631 IB_EVENT_SQ_DRAINED, 632 IB_EVENT_PATH_MIG, 633 IB_EVENT_PATH_MIG_ERR, 634 IB_EVENT_DEVICE_FATAL, 635 IB_EVENT_PORT_ACTIVE, 636 IB_EVENT_PORT_ERR, 637 IB_EVENT_LID_CHANGE, 638 IB_EVENT_PKEY_CHANGE, 639 IB_EVENT_SM_CHANGE, 640 IB_EVENT_SRQ_ERR, 641 IB_EVENT_SRQ_LIMIT_REACHED, 642 IB_EVENT_QP_LAST_WQE_REACHED, 643 IB_EVENT_CLIENT_REREGISTER, 644 IB_EVENT_GID_CHANGE, 645 IB_EVENT_WQ_FATAL, 646 }; 647 648 const char *__attribute_const__ ib_event_msg(enum ib_event_type event); 649 650 struct ib_event { 651 struct ib_device *device; 652 union { 653 struct ib_cq *cq; 654 struct ib_qp *qp; 655 struct ib_srq *srq; 656 struct ib_wq *wq; 657 u8 port_num; 658 } element; 659 enum ib_event_type event; 660 }; 661 662 struct ib_event_handler { 663 struct ib_device *device; 664 void (*handler)(struct ib_event_handler *, struct ib_event *); 665 struct list_head list; 666 }; 667 668 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \ 669 do { \ 670 (_ptr)->device = _device; \ 671 (_ptr)->handler = _handler; \ 672 INIT_LIST_HEAD(&(_ptr)->list); \ 673 } while (0) 674 675 struct ib_global_route { 676 const struct ib_gid_attr *sgid_attr; 677 union ib_gid dgid; 678 u32 flow_label; 679 u8 sgid_index; 680 u8 hop_limit; 681 u8 traffic_class; 682 }; 683 684 struct ib_grh { 685 __be32 version_tclass_flow; 686 __be16 paylen; 687 u8 next_hdr; 688 u8 hop_limit; 689 union ib_gid sgid; 690 union ib_gid dgid; 691 }; 692 693 union rdma_network_hdr { 694 struct ib_grh ibgrh; 695 struct { 696 /* The IB spec states that if it's IPv4, the header 697 * is located in the last 20 bytes of the header. 698 */ 699 u8 reserved[20]; 700 struct iphdr roce4grh; 701 }; 702 }; 703 704 #define IB_QPN_MASK 0xFFFFFF 705 706 enum { 707 IB_MULTICAST_QPN = 0xffffff 708 }; 709 710 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF) 711 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000) 712 713 enum ib_ah_flags { 714 IB_AH_GRH = 1 715 }; 716 717 enum ib_rate { 718 IB_RATE_PORT_CURRENT = 0, 719 IB_RATE_2_5_GBPS = 2, 720 IB_RATE_5_GBPS = 5, 721 IB_RATE_10_GBPS = 3, 722 IB_RATE_20_GBPS = 6, 723 IB_RATE_30_GBPS = 4, 724 IB_RATE_40_GBPS = 7, 725 IB_RATE_60_GBPS = 8, 726 IB_RATE_80_GBPS = 9, 727 IB_RATE_120_GBPS = 10, 728 IB_RATE_14_GBPS = 11, 729 IB_RATE_56_GBPS = 12, 730 IB_RATE_112_GBPS = 13, 731 IB_RATE_168_GBPS = 14, 732 IB_RATE_25_GBPS = 15, 733 IB_RATE_100_GBPS = 16, 734 IB_RATE_200_GBPS = 17, 735 IB_RATE_300_GBPS = 18, 736 IB_RATE_28_GBPS = 19, 737 IB_RATE_50_GBPS = 20, 738 IB_RATE_400_GBPS = 21, 739 IB_RATE_600_GBPS = 22, 740 }; 741 742 /** 743 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the 744 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be 745 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec. 746 * @rate: rate to convert. 747 */ 748 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate); 749 750 /** 751 * ib_rate_to_mbps - Convert the IB rate enum to Mbps. 752 * For example, IB_RATE_2_5_GBPS will be converted to 2500. 753 * @rate: rate to convert. 754 */ 755 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate); 756 757 758 /** 759 * enum ib_mr_type - memory region type 760 * @IB_MR_TYPE_MEM_REG: memory region that is used for 761 * normal registration 762 * @IB_MR_TYPE_SIGNATURE: memory region that is used for 763 * signature operations (data-integrity 764 * capable regions) 765 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to 766 * register any arbitrary sg lists (without 767 * the normal mr constraints - see 768 * ib_map_mr_sg) 769 */ 770 enum ib_mr_type { 771 IB_MR_TYPE_MEM_REG, 772 IB_MR_TYPE_SIGNATURE, 773 IB_MR_TYPE_SG_GAPS, 774 }; 775 776 /** 777 * Signature types 778 * IB_SIG_TYPE_NONE: Unprotected. 779 * IB_SIG_TYPE_T10_DIF: Type T10-DIF 780 */ 781 enum ib_signature_type { 782 IB_SIG_TYPE_NONE, 783 IB_SIG_TYPE_T10_DIF, 784 }; 785 786 /** 787 * Signature T10-DIF block-guard types 788 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules. 789 * IB_T10DIF_CSUM: Corresponds to IP checksum rules. 790 */ 791 enum ib_t10_dif_bg_type { 792 IB_T10DIF_CRC, 793 IB_T10DIF_CSUM 794 }; 795 796 /** 797 * struct ib_t10_dif_domain - Parameters specific for T10-DIF 798 * domain. 799 * @bg_type: T10-DIF block guard type (CRC|CSUM) 800 * @pi_interval: protection information interval. 801 * @bg: seed of guard computation. 802 * @app_tag: application tag of guard block 803 * @ref_tag: initial guard block reference tag. 804 * @ref_remap: Indicate wethear the reftag increments each block 805 * @app_escape: Indicate to skip block check if apptag=0xffff 806 * @ref_escape: Indicate to skip block check if reftag=0xffffffff 807 * @apptag_check_mask: check bitmask of application tag. 808 */ 809 struct ib_t10_dif_domain { 810 enum ib_t10_dif_bg_type bg_type; 811 u16 pi_interval; 812 u16 bg; 813 u16 app_tag; 814 u32 ref_tag; 815 bool ref_remap; 816 bool app_escape; 817 bool ref_escape; 818 u16 apptag_check_mask; 819 }; 820 821 /** 822 * struct ib_sig_domain - Parameters for signature domain 823 * @sig_type: specific signauture type 824 * @sig: union of all signature domain attributes that may 825 * be used to set domain layout. 826 */ 827 struct ib_sig_domain { 828 enum ib_signature_type sig_type; 829 union { 830 struct ib_t10_dif_domain dif; 831 } sig; 832 }; 833 834 /** 835 * struct ib_sig_attrs - Parameters for signature handover operation 836 * @check_mask: bitmask for signature byte check (8 bytes) 837 * @mem: memory domain layout desciptor. 838 * @wire: wire domain layout desciptor. 839 */ 840 struct ib_sig_attrs { 841 u8 check_mask; 842 struct ib_sig_domain mem; 843 struct ib_sig_domain wire; 844 }; 845 846 enum ib_sig_err_type { 847 IB_SIG_BAD_GUARD, 848 IB_SIG_BAD_REFTAG, 849 IB_SIG_BAD_APPTAG, 850 }; 851 852 /** 853 * Signature check masks (8 bytes in total) according to the T10-PI standard: 854 * -------- -------- ------------ 855 * | GUARD | APPTAG | REFTAG | 856 * | 2B | 2B | 4B | 857 * -------- -------- ------------ 858 */ 859 enum { 860 IB_SIG_CHECK_GUARD = 0xc0, 861 IB_SIG_CHECK_APPTAG = 0x30, 862 IB_SIG_CHECK_REFTAG = 0x0f, 863 }; 864 865 /** 866 * struct ib_sig_err - signature error descriptor 867 */ 868 struct ib_sig_err { 869 enum ib_sig_err_type err_type; 870 u32 expected; 871 u32 actual; 872 u64 sig_err_offset; 873 u32 key; 874 }; 875 876 enum ib_mr_status_check { 877 IB_MR_CHECK_SIG_STATUS = 1, 878 }; 879 880 /** 881 * struct ib_mr_status - Memory region status container 882 * 883 * @fail_status: Bitmask of MR checks status. For each 884 * failed check a corresponding status bit is set. 885 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS 886 * failure. 887 */ 888 struct ib_mr_status { 889 u32 fail_status; 890 struct ib_sig_err sig_err; 891 }; 892 893 /** 894 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate 895 * enum. 896 * @mult: multiple to convert. 897 */ 898 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult); 899 900 enum rdma_ah_attr_type { 901 RDMA_AH_ATTR_TYPE_UNDEFINED, 902 RDMA_AH_ATTR_TYPE_IB, 903 RDMA_AH_ATTR_TYPE_ROCE, 904 RDMA_AH_ATTR_TYPE_OPA, 905 }; 906 907 struct ib_ah_attr { 908 u16 dlid; 909 u8 src_path_bits; 910 }; 911 912 struct roce_ah_attr { 913 u8 dmac[ETH_ALEN]; 914 }; 915 916 struct opa_ah_attr { 917 u32 dlid; 918 u8 src_path_bits; 919 bool make_grd; 920 }; 921 922 struct rdma_ah_attr { 923 struct ib_global_route grh; 924 u8 sl; 925 u8 static_rate; 926 u8 port_num; 927 u8 ah_flags; 928 enum rdma_ah_attr_type type; 929 union { 930 struct ib_ah_attr ib; 931 struct roce_ah_attr roce; 932 struct opa_ah_attr opa; 933 }; 934 }; 935 936 enum ib_wc_status { 937 IB_WC_SUCCESS, 938 IB_WC_LOC_LEN_ERR, 939 IB_WC_LOC_QP_OP_ERR, 940 IB_WC_LOC_EEC_OP_ERR, 941 IB_WC_LOC_PROT_ERR, 942 IB_WC_WR_FLUSH_ERR, 943 IB_WC_MW_BIND_ERR, 944 IB_WC_BAD_RESP_ERR, 945 IB_WC_LOC_ACCESS_ERR, 946 IB_WC_REM_INV_REQ_ERR, 947 IB_WC_REM_ACCESS_ERR, 948 IB_WC_REM_OP_ERR, 949 IB_WC_RETRY_EXC_ERR, 950 IB_WC_RNR_RETRY_EXC_ERR, 951 IB_WC_LOC_RDD_VIOL_ERR, 952 IB_WC_REM_INV_RD_REQ_ERR, 953 IB_WC_REM_ABORT_ERR, 954 IB_WC_INV_EECN_ERR, 955 IB_WC_INV_EEC_STATE_ERR, 956 IB_WC_FATAL_ERR, 957 IB_WC_RESP_TIMEOUT_ERR, 958 IB_WC_GENERAL_ERR 959 }; 960 961 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status); 962 963 enum ib_wc_opcode { 964 IB_WC_SEND, 965 IB_WC_RDMA_WRITE, 966 IB_WC_RDMA_READ, 967 IB_WC_COMP_SWAP, 968 IB_WC_FETCH_ADD, 969 IB_WC_LSO, 970 IB_WC_LOCAL_INV, 971 IB_WC_REG_MR, 972 IB_WC_MASKED_COMP_SWAP, 973 IB_WC_MASKED_FETCH_ADD, 974 /* 975 * Set value of IB_WC_RECV so consumers can test if a completion is a 976 * receive by testing (opcode & IB_WC_RECV). 977 */ 978 IB_WC_RECV = 1 << 7, 979 IB_WC_RECV_RDMA_WITH_IMM 980 }; 981 982 enum ib_wc_flags { 983 IB_WC_GRH = 1, 984 IB_WC_WITH_IMM = (1<<1), 985 IB_WC_WITH_INVALIDATE = (1<<2), 986 IB_WC_IP_CSUM_OK = (1<<3), 987 IB_WC_WITH_SMAC = (1<<4), 988 IB_WC_WITH_VLAN = (1<<5), 989 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6), 990 }; 991 992 struct ib_wc { 993 union { 994 u64 wr_id; 995 struct ib_cqe *wr_cqe; 996 }; 997 enum ib_wc_status status; 998 enum ib_wc_opcode opcode; 999 u32 vendor_err; 1000 u32 byte_len; 1001 struct ib_qp *qp; 1002 union { 1003 __be32 imm_data; 1004 u32 invalidate_rkey; 1005 } ex; 1006 u32 src_qp; 1007 u32 slid; 1008 int wc_flags; 1009 u16 pkey_index; 1010 u8 sl; 1011 u8 dlid_path_bits; 1012 u8 port_num; /* valid only for DR SMPs on switches */ 1013 u8 smac[ETH_ALEN]; 1014 u16 vlan_id; 1015 u8 network_hdr_type; 1016 }; 1017 1018 enum ib_cq_notify_flags { 1019 IB_CQ_SOLICITED = 1 << 0, 1020 IB_CQ_NEXT_COMP = 1 << 1, 1021 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP, 1022 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2, 1023 }; 1024 1025 enum ib_srq_type { 1026 IB_SRQT_BASIC, 1027 IB_SRQT_XRC, 1028 IB_SRQT_TM, 1029 }; 1030 1031 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type) 1032 { 1033 return srq_type == IB_SRQT_XRC || 1034 srq_type == IB_SRQT_TM; 1035 } 1036 1037 enum ib_srq_attr_mask { 1038 IB_SRQ_MAX_WR = 1 << 0, 1039 IB_SRQ_LIMIT = 1 << 1, 1040 }; 1041 1042 struct ib_srq_attr { 1043 u32 max_wr; 1044 u32 max_sge; 1045 u32 srq_limit; 1046 }; 1047 1048 struct ib_srq_init_attr { 1049 void (*event_handler)(struct ib_event *, void *); 1050 void *srq_context; 1051 struct ib_srq_attr attr; 1052 enum ib_srq_type srq_type; 1053 1054 struct { 1055 struct ib_cq *cq; 1056 union { 1057 struct { 1058 struct ib_xrcd *xrcd; 1059 } xrc; 1060 1061 struct { 1062 u32 max_num_tags; 1063 } tag_matching; 1064 }; 1065 } ext; 1066 }; 1067 1068 struct ib_qp_cap { 1069 u32 max_send_wr; 1070 u32 max_recv_wr; 1071 u32 max_send_sge; 1072 u32 max_recv_sge; 1073 u32 max_inline_data; 1074 1075 /* 1076 * Maximum number of rdma_rw_ctx structures in flight at a time. 1077 * ib_create_qp() will calculate the right amount of neededed WRs 1078 * and MRs based on this. 1079 */ 1080 u32 max_rdma_ctxs; 1081 }; 1082 1083 enum ib_sig_type { 1084 IB_SIGNAL_ALL_WR, 1085 IB_SIGNAL_REQ_WR 1086 }; 1087 1088 enum ib_qp_type { 1089 /* 1090 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries 1091 * here (and in that order) since the MAD layer uses them as 1092 * indices into a 2-entry table. 1093 */ 1094 IB_QPT_SMI, 1095 IB_QPT_GSI, 1096 1097 IB_QPT_RC, 1098 IB_QPT_UC, 1099 IB_QPT_UD, 1100 IB_QPT_RAW_IPV6, 1101 IB_QPT_RAW_ETHERTYPE, 1102 IB_QPT_RAW_PACKET = 8, 1103 IB_QPT_XRC_INI = 9, 1104 IB_QPT_XRC_TGT, 1105 IB_QPT_MAX, 1106 IB_QPT_DRIVER = 0xFF, 1107 /* Reserve a range for qp types internal to the low level driver. 1108 * These qp types will not be visible at the IB core layer, so the 1109 * IB_QPT_MAX usages should not be affected in the core layer 1110 */ 1111 IB_QPT_RESERVED1 = 0x1000, 1112 IB_QPT_RESERVED2, 1113 IB_QPT_RESERVED3, 1114 IB_QPT_RESERVED4, 1115 IB_QPT_RESERVED5, 1116 IB_QPT_RESERVED6, 1117 IB_QPT_RESERVED7, 1118 IB_QPT_RESERVED8, 1119 IB_QPT_RESERVED9, 1120 IB_QPT_RESERVED10, 1121 }; 1122 1123 enum ib_qp_create_flags { 1124 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0, 1125 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1, 1126 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2, 1127 IB_QP_CREATE_MANAGED_SEND = 1 << 3, 1128 IB_QP_CREATE_MANAGED_RECV = 1 << 4, 1129 IB_QP_CREATE_NETIF_QP = 1 << 5, 1130 IB_QP_CREATE_SIGNATURE_EN = 1 << 6, 1131 /* FREE = 1 << 7, */ 1132 IB_QP_CREATE_SCATTER_FCS = 1 << 8, 1133 IB_QP_CREATE_CVLAN_STRIPPING = 1 << 9, 1134 IB_QP_CREATE_SOURCE_QPN = 1 << 10, 1135 IB_QP_CREATE_PCI_WRITE_END_PADDING = 1 << 11, 1136 /* reserve bits 26-31 for low level drivers' internal use */ 1137 IB_QP_CREATE_RESERVED_START = 1 << 26, 1138 IB_QP_CREATE_RESERVED_END = 1 << 31, 1139 }; 1140 1141 /* 1142 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler 1143 * callback to destroy the passed in QP. 1144 */ 1145 1146 struct ib_qp_init_attr { 1147 /* Consumer's event_handler callback must not block */ 1148 void (*event_handler)(struct ib_event *, void *); 1149 1150 void *qp_context; 1151 struct ib_cq *send_cq; 1152 struct ib_cq *recv_cq; 1153 struct ib_srq *srq; 1154 struct ib_xrcd *xrcd; /* XRC TGT QPs only */ 1155 struct ib_qp_cap cap; 1156 enum ib_sig_type sq_sig_type; 1157 enum ib_qp_type qp_type; 1158 u32 create_flags; 1159 1160 /* 1161 * Only needed for special QP types, or when using the RW API. 1162 */ 1163 u8 port_num; 1164 struct ib_rwq_ind_table *rwq_ind_tbl; 1165 u32 source_qpn; 1166 }; 1167 1168 struct ib_qp_open_attr { 1169 void (*event_handler)(struct ib_event *, void *); 1170 void *qp_context; 1171 u32 qp_num; 1172 enum ib_qp_type qp_type; 1173 }; 1174 1175 enum ib_rnr_timeout { 1176 IB_RNR_TIMER_655_36 = 0, 1177 IB_RNR_TIMER_000_01 = 1, 1178 IB_RNR_TIMER_000_02 = 2, 1179 IB_RNR_TIMER_000_03 = 3, 1180 IB_RNR_TIMER_000_04 = 4, 1181 IB_RNR_TIMER_000_06 = 5, 1182 IB_RNR_TIMER_000_08 = 6, 1183 IB_RNR_TIMER_000_12 = 7, 1184 IB_RNR_TIMER_000_16 = 8, 1185 IB_RNR_TIMER_000_24 = 9, 1186 IB_RNR_TIMER_000_32 = 10, 1187 IB_RNR_TIMER_000_48 = 11, 1188 IB_RNR_TIMER_000_64 = 12, 1189 IB_RNR_TIMER_000_96 = 13, 1190 IB_RNR_TIMER_001_28 = 14, 1191 IB_RNR_TIMER_001_92 = 15, 1192 IB_RNR_TIMER_002_56 = 16, 1193 IB_RNR_TIMER_003_84 = 17, 1194 IB_RNR_TIMER_005_12 = 18, 1195 IB_RNR_TIMER_007_68 = 19, 1196 IB_RNR_TIMER_010_24 = 20, 1197 IB_RNR_TIMER_015_36 = 21, 1198 IB_RNR_TIMER_020_48 = 22, 1199 IB_RNR_TIMER_030_72 = 23, 1200 IB_RNR_TIMER_040_96 = 24, 1201 IB_RNR_TIMER_061_44 = 25, 1202 IB_RNR_TIMER_081_92 = 26, 1203 IB_RNR_TIMER_122_88 = 27, 1204 IB_RNR_TIMER_163_84 = 28, 1205 IB_RNR_TIMER_245_76 = 29, 1206 IB_RNR_TIMER_327_68 = 30, 1207 IB_RNR_TIMER_491_52 = 31 1208 }; 1209 1210 enum ib_qp_attr_mask { 1211 IB_QP_STATE = 1, 1212 IB_QP_CUR_STATE = (1<<1), 1213 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2), 1214 IB_QP_ACCESS_FLAGS = (1<<3), 1215 IB_QP_PKEY_INDEX = (1<<4), 1216 IB_QP_PORT = (1<<5), 1217 IB_QP_QKEY = (1<<6), 1218 IB_QP_AV = (1<<7), 1219 IB_QP_PATH_MTU = (1<<8), 1220 IB_QP_TIMEOUT = (1<<9), 1221 IB_QP_RETRY_CNT = (1<<10), 1222 IB_QP_RNR_RETRY = (1<<11), 1223 IB_QP_RQ_PSN = (1<<12), 1224 IB_QP_MAX_QP_RD_ATOMIC = (1<<13), 1225 IB_QP_ALT_PATH = (1<<14), 1226 IB_QP_MIN_RNR_TIMER = (1<<15), 1227 IB_QP_SQ_PSN = (1<<16), 1228 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17), 1229 IB_QP_PATH_MIG_STATE = (1<<18), 1230 IB_QP_CAP = (1<<19), 1231 IB_QP_DEST_QPN = (1<<20), 1232 IB_QP_RESERVED1 = (1<<21), 1233 IB_QP_RESERVED2 = (1<<22), 1234 IB_QP_RESERVED3 = (1<<23), 1235 IB_QP_RESERVED4 = (1<<24), 1236 IB_QP_RATE_LIMIT = (1<<25), 1237 }; 1238 1239 enum ib_qp_state { 1240 IB_QPS_RESET, 1241 IB_QPS_INIT, 1242 IB_QPS_RTR, 1243 IB_QPS_RTS, 1244 IB_QPS_SQD, 1245 IB_QPS_SQE, 1246 IB_QPS_ERR 1247 }; 1248 1249 enum ib_mig_state { 1250 IB_MIG_MIGRATED, 1251 IB_MIG_REARM, 1252 IB_MIG_ARMED 1253 }; 1254 1255 enum ib_mw_type { 1256 IB_MW_TYPE_1 = 1, 1257 IB_MW_TYPE_2 = 2 1258 }; 1259 1260 struct ib_qp_attr { 1261 enum ib_qp_state qp_state; 1262 enum ib_qp_state cur_qp_state; 1263 enum ib_mtu path_mtu; 1264 enum ib_mig_state path_mig_state; 1265 u32 qkey; 1266 u32 rq_psn; 1267 u32 sq_psn; 1268 u32 dest_qp_num; 1269 int qp_access_flags; 1270 struct ib_qp_cap cap; 1271 struct rdma_ah_attr ah_attr; 1272 struct rdma_ah_attr alt_ah_attr; 1273 u16 pkey_index; 1274 u16 alt_pkey_index; 1275 u8 en_sqd_async_notify; 1276 u8 sq_draining; 1277 u8 max_rd_atomic; 1278 u8 max_dest_rd_atomic; 1279 u8 min_rnr_timer; 1280 u8 port_num; 1281 u8 timeout; 1282 u8 retry_cnt; 1283 u8 rnr_retry; 1284 u8 alt_port_num; 1285 u8 alt_timeout; 1286 u32 rate_limit; 1287 }; 1288 1289 enum ib_wr_opcode { 1290 /* These are shared with userspace */ 1291 IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE, 1292 IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM, 1293 IB_WR_SEND = IB_UVERBS_WR_SEND, 1294 IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM, 1295 IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ, 1296 IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP, 1297 IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD, 1298 IB_WR_LSO = IB_UVERBS_WR_TSO, 1299 IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV, 1300 IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV, 1301 IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV, 1302 IB_WR_MASKED_ATOMIC_CMP_AND_SWP = 1303 IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP, 1304 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD = 1305 IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD, 1306 1307 /* These are kernel only and can not be issued by userspace */ 1308 IB_WR_REG_MR = 0x20, 1309 IB_WR_REG_SIG_MR, 1310 1311 /* reserve values for low level drivers' internal use. 1312 * These values will not be used at all in the ib core layer. 1313 */ 1314 IB_WR_RESERVED1 = 0xf0, 1315 IB_WR_RESERVED2, 1316 IB_WR_RESERVED3, 1317 IB_WR_RESERVED4, 1318 IB_WR_RESERVED5, 1319 IB_WR_RESERVED6, 1320 IB_WR_RESERVED7, 1321 IB_WR_RESERVED8, 1322 IB_WR_RESERVED9, 1323 IB_WR_RESERVED10, 1324 }; 1325 1326 enum ib_send_flags { 1327 IB_SEND_FENCE = 1, 1328 IB_SEND_SIGNALED = (1<<1), 1329 IB_SEND_SOLICITED = (1<<2), 1330 IB_SEND_INLINE = (1<<3), 1331 IB_SEND_IP_CSUM = (1<<4), 1332 1333 /* reserve bits 26-31 for low level drivers' internal use */ 1334 IB_SEND_RESERVED_START = (1 << 26), 1335 IB_SEND_RESERVED_END = (1 << 31), 1336 }; 1337 1338 struct ib_sge { 1339 u64 addr; 1340 u32 length; 1341 u32 lkey; 1342 }; 1343 1344 struct ib_cqe { 1345 void (*done)(struct ib_cq *cq, struct ib_wc *wc); 1346 }; 1347 1348 struct ib_send_wr { 1349 struct ib_send_wr *next; 1350 union { 1351 u64 wr_id; 1352 struct ib_cqe *wr_cqe; 1353 }; 1354 struct ib_sge *sg_list; 1355 int num_sge; 1356 enum ib_wr_opcode opcode; 1357 int send_flags; 1358 union { 1359 __be32 imm_data; 1360 u32 invalidate_rkey; 1361 } ex; 1362 }; 1363 1364 struct ib_rdma_wr { 1365 struct ib_send_wr wr; 1366 u64 remote_addr; 1367 u32 rkey; 1368 }; 1369 1370 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr) 1371 { 1372 return container_of(wr, struct ib_rdma_wr, wr); 1373 } 1374 1375 struct ib_atomic_wr { 1376 struct ib_send_wr wr; 1377 u64 remote_addr; 1378 u64 compare_add; 1379 u64 swap; 1380 u64 compare_add_mask; 1381 u64 swap_mask; 1382 u32 rkey; 1383 }; 1384 1385 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr) 1386 { 1387 return container_of(wr, struct ib_atomic_wr, wr); 1388 } 1389 1390 struct ib_ud_wr { 1391 struct ib_send_wr wr; 1392 struct ib_ah *ah; 1393 void *header; 1394 int hlen; 1395 int mss; 1396 u32 remote_qpn; 1397 u32 remote_qkey; 1398 u16 pkey_index; /* valid for GSI only */ 1399 u8 port_num; /* valid for DR SMPs on switch only */ 1400 }; 1401 1402 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr) 1403 { 1404 return container_of(wr, struct ib_ud_wr, wr); 1405 } 1406 1407 struct ib_reg_wr { 1408 struct ib_send_wr wr; 1409 struct ib_mr *mr; 1410 u32 key; 1411 int access; 1412 }; 1413 1414 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr) 1415 { 1416 return container_of(wr, struct ib_reg_wr, wr); 1417 } 1418 1419 struct ib_sig_handover_wr { 1420 struct ib_send_wr wr; 1421 struct ib_sig_attrs *sig_attrs; 1422 struct ib_mr *sig_mr; 1423 int access_flags; 1424 struct ib_sge *prot; 1425 }; 1426 1427 static inline const struct ib_sig_handover_wr * 1428 sig_handover_wr(const struct ib_send_wr *wr) 1429 { 1430 return container_of(wr, struct ib_sig_handover_wr, wr); 1431 } 1432 1433 struct ib_recv_wr { 1434 struct ib_recv_wr *next; 1435 union { 1436 u64 wr_id; 1437 struct ib_cqe *wr_cqe; 1438 }; 1439 struct ib_sge *sg_list; 1440 int num_sge; 1441 }; 1442 1443 enum ib_access_flags { 1444 IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE, 1445 IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE, 1446 IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ, 1447 IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC, 1448 IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND, 1449 IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED, 1450 IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND, 1451 IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB, 1452 1453 IB_ACCESS_SUPPORTED = ((IB_ACCESS_HUGETLB << 1) - 1) 1454 }; 1455 1456 /* 1457 * XXX: these are apparently used for ->rereg_user_mr, no idea why they 1458 * are hidden here instead of a uapi header! 1459 */ 1460 enum ib_mr_rereg_flags { 1461 IB_MR_REREG_TRANS = 1, 1462 IB_MR_REREG_PD = (1<<1), 1463 IB_MR_REREG_ACCESS = (1<<2), 1464 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1) 1465 }; 1466 1467 struct ib_fmr_attr { 1468 int max_pages; 1469 int max_maps; 1470 u8 page_shift; 1471 }; 1472 1473 struct ib_umem; 1474 1475 enum rdma_remove_reason { 1476 /* 1477 * Userspace requested uobject deletion or initial try 1478 * to remove uobject via cleanup. Call could fail 1479 */ 1480 RDMA_REMOVE_DESTROY, 1481 /* Context deletion. This call should delete the actual object itself */ 1482 RDMA_REMOVE_CLOSE, 1483 /* Driver is being hot-unplugged. This call should delete the actual object itself */ 1484 RDMA_REMOVE_DRIVER_REMOVE, 1485 /* uobj is being cleaned-up before being committed */ 1486 RDMA_REMOVE_ABORT, 1487 }; 1488 1489 struct ib_rdmacg_object { 1490 #ifdef CONFIG_CGROUP_RDMA 1491 struct rdma_cgroup *cg; /* owner rdma cgroup */ 1492 #endif 1493 }; 1494 1495 struct ib_ucontext { 1496 struct ib_device *device; 1497 struct ib_uverbs_file *ufile; 1498 /* 1499 * 'closing' can be read by the driver only during a destroy callback, 1500 * it is set when we are closing the file descriptor and indicates 1501 * that mm_sem may be locked. 1502 */ 1503 bool closing; 1504 1505 bool cleanup_retryable; 1506 1507 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING 1508 void (*invalidate_range)(struct ib_umem_odp *umem_odp, 1509 unsigned long start, unsigned long end); 1510 struct mutex per_mm_list_lock; 1511 struct list_head per_mm_list; 1512 #endif 1513 1514 struct ib_rdmacg_object cg_obj; 1515 /* 1516 * Implementation details of the RDMA core, don't use in drivers: 1517 */ 1518 struct rdma_restrack_entry res; 1519 }; 1520 1521 struct ib_uobject { 1522 u64 user_handle; /* handle given to us by userspace */ 1523 /* ufile & ucontext owning this object */ 1524 struct ib_uverbs_file *ufile; 1525 /* FIXME, save memory: ufile->context == context */ 1526 struct ib_ucontext *context; /* associated user context */ 1527 void *object; /* containing object */ 1528 struct list_head list; /* link to context's list */ 1529 struct ib_rdmacg_object cg_obj; /* rdmacg object */ 1530 int id; /* index into kernel idr */ 1531 struct kref ref; 1532 atomic_t usecnt; /* protects exclusive access */ 1533 struct rcu_head rcu; /* kfree_rcu() overhead */ 1534 1535 const struct uverbs_api_object *uapi_object; 1536 }; 1537 1538 struct ib_udata { 1539 const void __user *inbuf; 1540 void __user *outbuf; 1541 size_t inlen; 1542 size_t outlen; 1543 }; 1544 1545 struct ib_pd { 1546 u32 local_dma_lkey; 1547 u32 flags; 1548 struct ib_device *device; 1549 struct ib_uobject *uobject; 1550 atomic_t usecnt; /* count all resources */ 1551 1552 u32 unsafe_global_rkey; 1553 1554 /* 1555 * Implementation details of the RDMA core, don't use in drivers: 1556 */ 1557 struct ib_mr *__internal_mr; 1558 struct rdma_restrack_entry res; 1559 }; 1560 1561 struct ib_xrcd { 1562 struct ib_device *device; 1563 atomic_t usecnt; /* count all exposed resources */ 1564 struct inode *inode; 1565 1566 struct mutex tgt_qp_mutex; 1567 struct list_head tgt_qp_list; 1568 }; 1569 1570 struct ib_ah { 1571 struct ib_device *device; 1572 struct ib_pd *pd; 1573 struct ib_uobject *uobject; 1574 const struct ib_gid_attr *sgid_attr; 1575 enum rdma_ah_attr_type type; 1576 }; 1577 1578 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context); 1579 1580 enum ib_poll_context { 1581 IB_POLL_DIRECT, /* caller context, no hw completions */ 1582 IB_POLL_SOFTIRQ, /* poll from softirq context */ 1583 IB_POLL_WORKQUEUE, /* poll from workqueue */ 1584 IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */ 1585 }; 1586 1587 struct ib_cq { 1588 struct ib_device *device; 1589 struct ib_uobject *uobject; 1590 ib_comp_handler comp_handler; 1591 void (*event_handler)(struct ib_event *, void *); 1592 void *cq_context; 1593 int cqe; 1594 atomic_t usecnt; /* count number of work queues */ 1595 enum ib_poll_context poll_ctx; 1596 struct ib_wc *wc; 1597 union { 1598 struct irq_poll iop; 1599 struct work_struct work; 1600 }; 1601 struct workqueue_struct *comp_wq; 1602 /* 1603 * Implementation details of the RDMA core, don't use in drivers: 1604 */ 1605 struct rdma_restrack_entry res; 1606 }; 1607 1608 struct ib_srq { 1609 struct ib_device *device; 1610 struct ib_pd *pd; 1611 struct ib_uobject *uobject; 1612 void (*event_handler)(struct ib_event *, void *); 1613 void *srq_context; 1614 enum ib_srq_type srq_type; 1615 atomic_t usecnt; 1616 1617 struct { 1618 struct ib_cq *cq; 1619 union { 1620 struct { 1621 struct ib_xrcd *xrcd; 1622 u32 srq_num; 1623 } xrc; 1624 }; 1625 } ext; 1626 }; 1627 1628 enum ib_raw_packet_caps { 1629 /* Strip cvlan from incoming packet and report it in the matching work 1630 * completion is supported. 1631 */ 1632 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = (1 << 0), 1633 /* Scatter FCS field of an incoming packet to host memory is supported. 1634 */ 1635 IB_RAW_PACKET_CAP_SCATTER_FCS = (1 << 1), 1636 /* Checksum offloads are supported (for both send and receive). */ 1637 IB_RAW_PACKET_CAP_IP_CSUM = (1 << 2), 1638 /* When a packet is received for an RQ with no receive WQEs, the 1639 * packet processing is delayed. 1640 */ 1641 IB_RAW_PACKET_CAP_DELAY_DROP = (1 << 3), 1642 }; 1643 1644 enum ib_wq_type { 1645 IB_WQT_RQ 1646 }; 1647 1648 enum ib_wq_state { 1649 IB_WQS_RESET, 1650 IB_WQS_RDY, 1651 IB_WQS_ERR 1652 }; 1653 1654 struct ib_wq { 1655 struct ib_device *device; 1656 struct ib_uobject *uobject; 1657 void *wq_context; 1658 void (*event_handler)(struct ib_event *, void *); 1659 struct ib_pd *pd; 1660 struct ib_cq *cq; 1661 u32 wq_num; 1662 enum ib_wq_state state; 1663 enum ib_wq_type wq_type; 1664 atomic_t usecnt; 1665 }; 1666 1667 enum ib_wq_flags { 1668 IB_WQ_FLAGS_CVLAN_STRIPPING = 1 << 0, 1669 IB_WQ_FLAGS_SCATTER_FCS = 1 << 1, 1670 IB_WQ_FLAGS_DELAY_DROP = 1 << 2, 1671 IB_WQ_FLAGS_PCI_WRITE_END_PADDING = 1 << 3, 1672 }; 1673 1674 struct ib_wq_init_attr { 1675 void *wq_context; 1676 enum ib_wq_type wq_type; 1677 u32 max_wr; 1678 u32 max_sge; 1679 struct ib_cq *cq; 1680 void (*event_handler)(struct ib_event *, void *); 1681 u32 create_flags; /* Use enum ib_wq_flags */ 1682 }; 1683 1684 enum ib_wq_attr_mask { 1685 IB_WQ_STATE = 1 << 0, 1686 IB_WQ_CUR_STATE = 1 << 1, 1687 IB_WQ_FLAGS = 1 << 2, 1688 }; 1689 1690 struct ib_wq_attr { 1691 enum ib_wq_state wq_state; 1692 enum ib_wq_state curr_wq_state; 1693 u32 flags; /* Use enum ib_wq_flags */ 1694 u32 flags_mask; /* Use enum ib_wq_flags */ 1695 }; 1696 1697 struct ib_rwq_ind_table { 1698 struct ib_device *device; 1699 struct ib_uobject *uobject; 1700 atomic_t usecnt; 1701 u32 ind_tbl_num; 1702 u32 log_ind_tbl_size; 1703 struct ib_wq **ind_tbl; 1704 }; 1705 1706 struct ib_rwq_ind_table_init_attr { 1707 u32 log_ind_tbl_size; 1708 /* Each entry is a pointer to Receive Work Queue */ 1709 struct ib_wq **ind_tbl; 1710 }; 1711 1712 enum port_pkey_state { 1713 IB_PORT_PKEY_NOT_VALID = 0, 1714 IB_PORT_PKEY_VALID = 1, 1715 IB_PORT_PKEY_LISTED = 2, 1716 }; 1717 1718 struct ib_qp_security; 1719 1720 struct ib_port_pkey { 1721 enum port_pkey_state state; 1722 u16 pkey_index; 1723 u8 port_num; 1724 struct list_head qp_list; 1725 struct list_head to_error_list; 1726 struct ib_qp_security *sec; 1727 }; 1728 1729 struct ib_ports_pkeys { 1730 struct ib_port_pkey main; 1731 struct ib_port_pkey alt; 1732 }; 1733 1734 struct ib_qp_security { 1735 struct ib_qp *qp; 1736 struct ib_device *dev; 1737 /* Hold this mutex when changing port and pkey settings. */ 1738 struct mutex mutex; 1739 struct ib_ports_pkeys *ports_pkeys; 1740 /* A list of all open shared QP handles. Required to enforce security 1741 * properly for all users of a shared QP. 1742 */ 1743 struct list_head shared_qp_list; 1744 void *security; 1745 bool destroying; 1746 atomic_t error_list_count; 1747 struct completion error_complete; 1748 int error_comps_pending; 1749 }; 1750 1751 /* 1752 * @max_write_sge: Maximum SGE elements per RDMA WRITE request. 1753 * @max_read_sge: Maximum SGE elements per RDMA READ request. 1754 */ 1755 struct ib_qp { 1756 struct ib_device *device; 1757 struct ib_pd *pd; 1758 struct ib_cq *send_cq; 1759 struct ib_cq *recv_cq; 1760 spinlock_t mr_lock; 1761 int mrs_used; 1762 struct list_head rdma_mrs; 1763 struct list_head sig_mrs; 1764 struct ib_srq *srq; 1765 struct ib_xrcd *xrcd; /* XRC TGT QPs only */ 1766 struct list_head xrcd_list; 1767 1768 /* count times opened, mcast attaches, flow attaches */ 1769 atomic_t usecnt; 1770 struct list_head open_list; 1771 struct ib_qp *real_qp; 1772 struct ib_uobject *uobject; 1773 void (*event_handler)(struct ib_event *, void *); 1774 void *qp_context; 1775 /* sgid_attrs associated with the AV's */ 1776 const struct ib_gid_attr *av_sgid_attr; 1777 const struct ib_gid_attr *alt_path_sgid_attr; 1778 u32 qp_num; 1779 u32 max_write_sge; 1780 u32 max_read_sge; 1781 enum ib_qp_type qp_type; 1782 struct ib_rwq_ind_table *rwq_ind_tbl; 1783 struct ib_qp_security *qp_sec; 1784 u8 port; 1785 1786 /* 1787 * Implementation details of the RDMA core, don't use in drivers: 1788 */ 1789 struct rdma_restrack_entry res; 1790 }; 1791 1792 struct ib_dm { 1793 struct ib_device *device; 1794 u32 length; 1795 u32 flags; 1796 struct ib_uobject *uobject; 1797 atomic_t usecnt; 1798 }; 1799 1800 struct ib_mr { 1801 struct ib_device *device; 1802 struct ib_pd *pd; 1803 u32 lkey; 1804 u32 rkey; 1805 u64 iova; 1806 u64 length; 1807 unsigned int page_size; 1808 bool need_inval; 1809 union { 1810 struct ib_uobject *uobject; /* user */ 1811 struct list_head qp_entry; /* FR */ 1812 }; 1813 1814 struct ib_dm *dm; 1815 1816 /* 1817 * Implementation details of the RDMA core, don't use in drivers: 1818 */ 1819 struct rdma_restrack_entry res; 1820 }; 1821 1822 struct ib_mw { 1823 struct ib_device *device; 1824 struct ib_pd *pd; 1825 struct ib_uobject *uobject; 1826 u32 rkey; 1827 enum ib_mw_type type; 1828 }; 1829 1830 struct ib_fmr { 1831 struct ib_device *device; 1832 struct ib_pd *pd; 1833 struct list_head list; 1834 u32 lkey; 1835 u32 rkey; 1836 }; 1837 1838 /* Supported steering options */ 1839 enum ib_flow_attr_type { 1840 /* steering according to rule specifications */ 1841 IB_FLOW_ATTR_NORMAL = 0x0, 1842 /* default unicast and multicast rule - 1843 * receive all Eth traffic which isn't steered to any QP 1844 */ 1845 IB_FLOW_ATTR_ALL_DEFAULT = 0x1, 1846 /* default multicast rule - 1847 * receive all Eth multicast traffic which isn't steered to any QP 1848 */ 1849 IB_FLOW_ATTR_MC_DEFAULT = 0x2, 1850 /* sniffer rule - receive all port traffic */ 1851 IB_FLOW_ATTR_SNIFFER = 0x3 1852 }; 1853 1854 /* Supported steering header types */ 1855 enum ib_flow_spec_type { 1856 /* L2 headers*/ 1857 IB_FLOW_SPEC_ETH = 0x20, 1858 IB_FLOW_SPEC_IB = 0x22, 1859 /* L3 header*/ 1860 IB_FLOW_SPEC_IPV4 = 0x30, 1861 IB_FLOW_SPEC_IPV6 = 0x31, 1862 IB_FLOW_SPEC_ESP = 0x34, 1863 /* L4 headers*/ 1864 IB_FLOW_SPEC_TCP = 0x40, 1865 IB_FLOW_SPEC_UDP = 0x41, 1866 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50, 1867 IB_FLOW_SPEC_GRE = 0x51, 1868 IB_FLOW_SPEC_MPLS = 0x60, 1869 IB_FLOW_SPEC_INNER = 0x100, 1870 /* Actions */ 1871 IB_FLOW_SPEC_ACTION_TAG = 0x1000, 1872 IB_FLOW_SPEC_ACTION_DROP = 0x1001, 1873 IB_FLOW_SPEC_ACTION_HANDLE = 0x1002, 1874 IB_FLOW_SPEC_ACTION_COUNT = 0x1003, 1875 }; 1876 #define IB_FLOW_SPEC_LAYER_MASK 0xF0 1877 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10 1878 1879 /* Flow steering rule priority is set according to it's domain. 1880 * Lower domain value means higher priority. 1881 */ 1882 enum ib_flow_domain { 1883 IB_FLOW_DOMAIN_USER, 1884 IB_FLOW_DOMAIN_ETHTOOL, 1885 IB_FLOW_DOMAIN_RFS, 1886 IB_FLOW_DOMAIN_NIC, 1887 IB_FLOW_DOMAIN_NUM /* Must be last */ 1888 }; 1889 1890 enum ib_flow_flags { 1891 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */ 1892 IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */ 1893 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 3 /* Must be last */ 1894 }; 1895 1896 struct ib_flow_eth_filter { 1897 u8 dst_mac[6]; 1898 u8 src_mac[6]; 1899 __be16 ether_type; 1900 __be16 vlan_tag; 1901 /* Must be last */ 1902 u8 real_sz[0]; 1903 }; 1904 1905 struct ib_flow_spec_eth { 1906 u32 type; 1907 u16 size; 1908 struct ib_flow_eth_filter val; 1909 struct ib_flow_eth_filter mask; 1910 }; 1911 1912 struct ib_flow_ib_filter { 1913 __be16 dlid; 1914 __u8 sl; 1915 /* Must be last */ 1916 u8 real_sz[0]; 1917 }; 1918 1919 struct ib_flow_spec_ib { 1920 u32 type; 1921 u16 size; 1922 struct ib_flow_ib_filter val; 1923 struct ib_flow_ib_filter mask; 1924 }; 1925 1926 /* IPv4 header flags */ 1927 enum ib_ipv4_flags { 1928 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */ 1929 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the 1930 last have this flag set */ 1931 }; 1932 1933 struct ib_flow_ipv4_filter { 1934 __be32 src_ip; 1935 __be32 dst_ip; 1936 u8 proto; 1937 u8 tos; 1938 u8 ttl; 1939 u8 flags; 1940 /* Must be last */ 1941 u8 real_sz[0]; 1942 }; 1943 1944 struct ib_flow_spec_ipv4 { 1945 u32 type; 1946 u16 size; 1947 struct ib_flow_ipv4_filter val; 1948 struct ib_flow_ipv4_filter mask; 1949 }; 1950 1951 struct ib_flow_ipv6_filter { 1952 u8 src_ip[16]; 1953 u8 dst_ip[16]; 1954 __be32 flow_label; 1955 u8 next_hdr; 1956 u8 traffic_class; 1957 u8 hop_limit; 1958 /* Must be last */ 1959 u8 real_sz[0]; 1960 }; 1961 1962 struct ib_flow_spec_ipv6 { 1963 u32 type; 1964 u16 size; 1965 struct ib_flow_ipv6_filter val; 1966 struct ib_flow_ipv6_filter mask; 1967 }; 1968 1969 struct ib_flow_tcp_udp_filter { 1970 __be16 dst_port; 1971 __be16 src_port; 1972 /* Must be last */ 1973 u8 real_sz[0]; 1974 }; 1975 1976 struct ib_flow_spec_tcp_udp { 1977 u32 type; 1978 u16 size; 1979 struct ib_flow_tcp_udp_filter val; 1980 struct ib_flow_tcp_udp_filter mask; 1981 }; 1982 1983 struct ib_flow_tunnel_filter { 1984 __be32 tunnel_id; 1985 u8 real_sz[0]; 1986 }; 1987 1988 /* ib_flow_spec_tunnel describes the Vxlan tunnel 1989 * the tunnel_id from val has the vni value 1990 */ 1991 struct ib_flow_spec_tunnel { 1992 u32 type; 1993 u16 size; 1994 struct ib_flow_tunnel_filter val; 1995 struct ib_flow_tunnel_filter mask; 1996 }; 1997 1998 struct ib_flow_esp_filter { 1999 __be32 spi; 2000 __be32 seq; 2001 /* Must be last */ 2002 u8 real_sz[0]; 2003 }; 2004 2005 struct ib_flow_spec_esp { 2006 u32 type; 2007 u16 size; 2008 struct ib_flow_esp_filter val; 2009 struct ib_flow_esp_filter mask; 2010 }; 2011 2012 struct ib_flow_gre_filter { 2013 __be16 c_ks_res0_ver; 2014 __be16 protocol; 2015 __be32 key; 2016 /* Must be last */ 2017 u8 real_sz[0]; 2018 }; 2019 2020 struct ib_flow_spec_gre { 2021 u32 type; 2022 u16 size; 2023 struct ib_flow_gre_filter val; 2024 struct ib_flow_gre_filter mask; 2025 }; 2026 2027 struct ib_flow_mpls_filter { 2028 __be32 tag; 2029 /* Must be last */ 2030 u8 real_sz[0]; 2031 }; 2032 2033 struct ib_flow_spec_mpls { 2034 u32 type; 2035 u16 size; 2036 struct ib_flow_mpls_filter val; 2037 struct ib_flow_mpls_filter mask; 2038 }; 2039 2040 struct ib_flow_spec_action_tag { 2041 enum ib_flow_spec_type type; 2042 u16 size; 2043 u32 tag_id; 2044 }; 2045 2046 struct ib_flow_spec_action_drop { 2047 enum ib_flow_spec_type type; 2048 u16 size; 2049 }; 2050 2051 struct ib_flow_spec_action_handle { 2052 enum ib_flow_spec_type type; 2053 u16 size; 2054 struct ib_flow_action *act; 2055 }; 2056 2057 enum ib_counters_description { 2058 IB_COUNTER_PACKETS, 2059 IB_COUNTER_BYTES, 2060 }; 2061 2062 struct ib_flow_spec_action_count { 2063 enum ib_flow_spec_type type; 2064 u16 size; 2065 struct ib_counters *counters; 2066 }; 2067 2068 union ib_flow_spec { 2069 struct { 2070 u32 type; 2071 u16 size; 2072 }; 2073 struct ib_flow_spec_eth eth; 2074 struct ib_flow_spec_ib ib; 2075 struct ib_flow_spec_ipv4 ipv4; 2076 struct ib_flow_spec_tcp_udp tcp_udp; 2077 struct ib_flow_spec_ipv6 ipv6; 2078 struct ib_flow_spec_tunnel tunnel; 2079 struct ib_flow_spec_esp esp; 2080 struct ib_flow_spec_gre gre; 2081 struct ib_flow_spec_mpls mpls; 2082 struct ib_flow_spec_action_tag flow_tag; 2083 struct ib_flow_spec_action_drop drop; 2084 struct ib_flow_spec_action_handle action; 2085 struct ib_flow_spec_action_count flow_count; 2086 }; 2087 2088 struct ib_flow_attr { 2089 enum ib_flow_attr_type type; 2090 u16 size; 2091 u16 priority; 2092 u32 flags; 2093 u8 num_of_specs; 2094 u8 port; 2095 union ib_flow_spec flows[]; 2096 }; 2097 2098 struct ib_flow { 2099 struct ib_qp *qp; 2100 struct ib_device *device; 2101 struct ib_uobject *uobject; 2102 }; 2103 2104 enum ib_flow_action_type { 2105 IB_FLOW_ACTION_UNSPECIFIED, 2106 IB_FLOW_ACTION_ESP = 1, 2107 }; 2108 2109 struct ib_flow_action_attrs_esp_keymats { 2110 enum ib_uverbs_flow_action_esp_keymat protocol; 2111 union { 2112 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm; 2113 } keymat; 2114 }; 2115 2116 struct ib_flow_action_attrs_esp_replays { 2117 enum ib_uverbs_flow_action_esp_replay protocol; 2118 union { 2119 struct ib_uverbs_flow_action_esp_replay_bmp bmp; 2120 } replay; 2121 }; 2122 2123 enum ib_flow_action_attrs_esp_flags { 2124 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags 2125 * This is done in order to share the same flags between user-space and 2126 * kernel and spare an unnecessary translation. 2127 */ 2128 2129 /* Kernel flags */ 2130 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32, 2131 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33, 2132 }; 2133 2134 struct ib_flow_spec_list { 2135 struct ib_flow_spec_list *next; 2136 union ib_flow_spec spec; 2137 }; 2138 2139 struct ib_flow_action_attrs_esp { 2140 struct ib_flow_action_attrs_esp_keymats *keymat; 2141 struct ib_flow_action_attrs_esp_replays *replay; 2142 struct ib_flow_spec_list *encap; 2143 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled. 2144 * Value of 0 is a valid value. 2145 */ 2146 u32 esn; 2147 u32 spi; 2148 u32 seq; 2149 u32 tfc_pad; 2150 /* Use enum ib_flow_action_attrs_esp_flags */ 2151 u64 flags; 2152 u64 hard_limit_pkts; 2153 }; 2154 2155 struct ib_flow_action { 2156 struct ib_device *device; 2157 struct ib_uobject *uobject; 2158 enum ib_flow_action_type type; 2159 atomic_t usecnt; 2160 }; 2161 2162 struct ib_mad_hdr; 2163 struct ib_grh; 2164 2165 enum ib_process_mad_flags { 2166 IB_MAD_IGNORE_MKEY = 1, 2167 IB_MAD_IGNORE_BKEY = 2, 2168 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY 2169 }; 2170 2171 enum ib_mad_result { 2172 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */ 2173 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */ 2174 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */ 2175 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */ 2176 }; 2177 2178 struct ib_port_cache { 2179 u64 subnet_prefix; 2180 struct ib_pkey_cache *pkey; 2181 struct ib_gid_table *gid; 2182 u8 lmc; 2183 enum ib_port_state port_state; 2184 }; 2185 2186 struct ib_cache { 2187 rwlock_t lock; 2188 struct ib_event_handler event_handler; 2189 struct ib_port_cache *ports; 2190 }; 2191 2192 struct iw_cm_verbs; 2193 2194 struct ib_port_immutable { 2195 int pkey_tbl_len; 2196 int gid_tbl_len; 2197 u32 core_cap_flags; 2198 u32 max_mad_size; 2199 }; 2200 2201 /* rdma netdev type - specifies protocol type */ 2202 enum rdma_netdev_t { 2203 RDMA_NETDEV_OPA_VNIC, 2204 RDMA_NETDEV_IPOIB, 2205 }; 2206 2207 /** 2208 * struct rdma_netdev - rdma netdev 2209 * For cases where netstack interfacing is required. 2210 */ 2211 struct rdma_netdev { 2212 void *clnt_priv; 2213 struct ib_device *hca; 2214 u8 port_num; 2215 2216 /* 2217 * cleanup function must be specified. 2218 * FIXME: This is only used for OPA_VNIC and that usage should be 2219 * removed too. 2220 */ 2221 void (*free_rdma_netdev)(struct net_device *netdev); 2222 2223 /* control functions */ 2224 void (*set_id)(struct net_device *netdev, int id); 2225 /* send packet */ 2226 int (*send)(struct net_device *dev, struct sk_buff *skb, 2227 struct ib_ah *address, u32 dqpn); 2228 /* multicast */ 2229 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca, 2230 union ib_gid *gid, u16 mlid, 2231 int set_qkey, u32 qkey); 2232 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca, 2233 union ib_gid *gid, u16 mlid); 2234 }; 2235 2236 struct rdma_netdev_alloc_params { 2237 size_t sizeof_priv; 2238 unsigned int txqs; 2239 unsigned int rxqs; 2240 void *param; 2241 2242 int (*initialize_rdma_netdev)(struct ib_device *device, u8 port_num, 2243 struct net_device *netdev, void *param); 2244 }; 2245 2246 struct ib_port_pkey_list { 2247 /* Lock to hold while modifying the list. */ 2248 spinlock_t list_lock; 2249 struct list_head pkey_list; 2250 }; 2251 2252 struct ib_counters { 2253 struct ib_device *device; 2254 struct ib_uobject *uobject; 2255 /* num of objects attached */ 2256 atomic_t usecnt; 2257 }; 2258 2259 struct ib_counters_read_attr { 2260 u64 *counters_buff; 2261 u32 ncounters; 2262 u32 flags; /* use enum ib_read_counters_flags */ 2263 }; 2264 2265 struct uverbs_attr_bundle; 2266 2267 /** 2268 * struct ib_device_ops - InfiniBand device operations 2269 * This structure defines all the InfiniBand device operations, providers will 2270 * need to define the supported operations, otherwise they will be set to null. 2271 */ 2272 struct ib_device_ops { 2273 int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr, 2274 const struct ib_send_wr **bad_send_wr); 2275 int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr, 2276 const struct ib_recv_wr **bad_recv_wr); 2277 void (*drain_rq)(struct ib_qp *qp); 2278 void (*drain_sq)(struct ib_qp *qp); 2279 int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc); 2280 int (*peek_cq)(struct ib_cq *cq, int wc_cnt); 2281 int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags); 2282 int (*req_ncomp_notif)(struct ib_cq *cq, int wc_cnt); 2283 int (*post_srq_recv)(struct ib_srq *srq, 2284 const struct ib_recv_wr *recv_wr, 2285 const struct ib_recv_wr **bad_recv_wr); 2286 int (*process_mad)(struct ib_device *device, int process_mad_flags, 2287 u8 port_num, const struct ib_wc *in_wc, 2288 const struct ib_grh *in_grh, 2289 const struct ib_mad_hdr *in_mad, size_t in_mad_size, 2290 struct ib_mad_hdr *out_mad, size_t *out_mad_size, 2291 u16 *out_mad_pkey_index); 2292 int (*query_device)(struct ib_device *device, 2293 struct ib_device_attr *device_attr, 2294 struct ib_udata *udata); 2295 int (*modify_device)(struct ib_device *device, int device_modify_mask, 2296 struct ib_device_modify *device_modify); 2297 void (*get_dev_fw_str)(struct ib_device *device, char *str); 2298 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev, 2299 int comp_vector); 2300 int (*query_port)(struct ib_device *device, u8 port_num, 2301 struct ib_port_attr *port_attr); 2302 int (*modify_port)(struct ib_device *device, u8 port_num, 2303 int port_modify_mask, 2304 struct ib_port_modify *port_modify); 2305 /** 2306 * The following mandatory functions are used only at device 2307 * registration. Keep functions such as these at the end of this 2308 * structure to avoid cache line misses when accessing struct ib_device 2309 * in fast paths. 2310 */ 2311 int (*get_port_immutable)(struct ib_device *device, u8 port_num, 2312 struct ib_port_immutable *immutable); 2313 enum rdma_link_layer (*get_link_layer)(struct ib_device *device, 2314 u8 port_num); 2315 /** 2316 * When calling get_netdev, the HW vendor's driver should return the 2317 * net device of device @device at port @port_num or NULL if such 2318 * a net device doesn't exist. The vendor driver should call dev_hold 2319 * on this net device. The HW vendor's device driver must guarantee 2320 * that this function returns NULL before the net device has finished 2321 * NETDEV_UNREGISTER state. 2322 */ 2323 struct net_device *(*get_netdev)(struct ib_device *device, u8 port_num); 2324 /** 2325 * rdma netdev operation 2326 * 2327 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params 2328 * must return -EOPNOTSUPP if it doesn't support the specified type. 2329 */ 2330 struct net_device *(*alloc_rdma_netdev)( 2331 struct ib_device *device, u8 port_num, enum rdma_netdev_t type, 2332 const char *name, unsigned char name_assign_type, 2333 void (*setup)(struct net_device *)); 2334 2335 int (*rdma_netdev_get_params)(struct ib_device *device, u8 port_num, 2336 enum rdma_netdev_t type, 2337 struct rdma_netdev_alloc_params *params); 2338 /** 2339 * query_gid should be return GID value for @device, when @port_num 2340 * link layer is either IB or iWarp. It is no-op if @port_num port 2341 * is RoCE link layer. 2342 */ 2343 int (*query_gid)(struct ib_device *device, u8 port_num, int index, 2344 union ib_gid *gid); 2345 /** 2346 * When calling add_gid, the HW vendor's driver should add the gid 2347 * of device of port at gid index available at @attr. Meta-info of 2348 * that gid (for example, the network device related to this gid) is 2349 * available at @attr. @context allows the HW vendor driver to store 2350 * extra information together with a GID entry. The HW vendor driver may 2351 * allocate memory to contain this information and store it in @context 2352 * when a new GID entry is written to. Params are consistent until the 2353 * next call of add_gid or delete_gid. The function should return 0 on 2354 * success or error otherwise. The function could be called 2355 * concurrently for different ports. This function is only called when 2356 * roce_gid_table is used. 2357 */ 2358 int (*add_gid)(const struct ib_gid_attr *attr, void **context); 2359 /** 2360 * When calling del_gid, the HW vendor's driver should delete the 2361 * gid of device @device at gid index gid_index of port port_num 2362 * available in @attr. 2363 * Upon the deletion of a GID entry, the HW vendor must free any 2364 * allocated memory. The caller will clear @context afterwards. 2365 * This function is only called when roce_gid_table is used. 2366 */ 2367 int (*del_gid)(const struct ib_gid_attr *attr, void **context); 2368 int (*query_pkey)(struct ib_device *device, u8 port_num, u16 index, 2369 u16 *pkey); 2370 struct ib_ucontext *(*alloc_ucontext)(struct ib_device *device, 2371 struct ib_udata *udata); 2372 int (*dealloc_ucontext)(struct ib_ucontext *context); 2373 int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma); 2374 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext); 2375 struct ib_pd *(*alloc_pd)(struct ib_device *device, 2376 struct ib_ucontext *context, 2377 struct ib_udata *udata); 2378 int (*dealloc_pd)(struct ib_pd *pd); 2379 struct ib_ah *(*create_ah)(struct ib_pd *pd, 2380 struct rdma_ah_attr *ah_attr, u32 flags, 2381 struct ib_udata *udata); 2382 int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr); 2383 int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr); 2384 int (*destroy_ah)(struct ib_ah *ah, u32 flags); 2385 struct ib_srq *(*create_srq)(struct ib_pd *pd, 2386 struct ib_srq_init_attr *srq_init_attr, 2387 struct ib_udata *udata); 2388 int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr, 2389 enum ib_srq_attr_mask srq_attr_mask, 2390 struct ib_udata *udata); 2391 int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr); 2392 int (*destroy_srq)(struct ib_srq *srq); 2393 struct ib_qp *(*create_qp)(struct ib_pd *pd, 2394 struct ib_qp_init_attr *qp_init_attr, 2395 struct ib_udata *udata); 2396 int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr, 2397 int qp_attr_mask, struct ib_udata *udata); 2398 int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr, 2399 int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr); 2400 int (*destroy_qp)(struct ib_qp *qp); 2401 struct ib_cq *(*create_cq)(struct ib_device *device, 2402 const struct ib_cq_init_attr *attr, 2403 struct ib_ucontext *context, 2404 struct ib_udata *udata); 2405 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period); 2406 int (*destroy_cq)(struct ib_cq *cq); 2407 int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata); 2408 struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags); 2409 struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length, 2410 u64 virt_addr, int mr_access_flags, 2411 struct ib_udata *udata); 2412 int (*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start, u64 length, 2413 u64 virt_addr, int mr_access_flags, 2414 struct ib_pd *pd, struct ib_udata *udata); 2415 int (*dereg_mr)(struct ib_mr *mr); 2416 struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type, 2417 u32 max_num_sg); 2418 int (*advise_mr)(struct ib_pd *pd, 2419 enum ib_uverbs_advise_mr_advice advice, u32 flags, 2420 struct ib_sge *sg_list, u32 num_sge, 2421 struct uverbs_attr_bundle *attrs); 2422 int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents, 2423 unsigned int *sg_offset); 2424 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask, 2425 struct ib_mr_status *mr_status); 2426 struct ib_mw *(*alloc_mw)(struct ib_pd *pd, enum ib_mw_type type, 2427 struct ib_udata *udata); 2428 int (*dealloc_mw)(struct ib_mw *mw); 2429 struct ib_fmr *(*alloc_fmr)(struct ib_pd *pd, int mr_access_flags, 2430 struct ib_fmr_attr *fmr_attr); 2431 int (*map_phys_fmr)(struct ib_fmr *fmr, u64 *page_list, int list_len, 2432 u64 iova); 2433 int (*unmap_fmr)(struct list_head *fmr_list); 2434 int (*dealloc_fmr)(struct ib_fmr *fmr); 2435 int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid); 2436 int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid); 2437 struct ib_xrcd *(*alloc_xrcd)(struct ib_device *device, 2438 struct ib_ucontext *ucontext, 2439 struct ib_udata *udata); 2440 int (*dealloc_xrcd)(struct ib_xrcd *xrcd); 2441 struct ib_flow *(*create_flow)(struct ib_qp *qp, 2442 struct ib_flow_attr *flow_attr, 2443 int domain, struct ib_udata *udata); 2444 int (*destroy_flow)(struct ib_flow *flow_id); 2445 struct ib_flow_action *(*create_flow_action_esp)( 2446 struct ib_device *device, 2447 const struct ib_flow_action_attrs_esp *attr, 2448 struct uverbs_attr_bundle *attrs); 2449 int (*destroy_flow_action)(struct ib_flow_action *action); 2450 int (*modify_flow_action_esp)( 2451 struct ib_flow_action *action, 2452 const struct ib_flow_action_attrs_esp *attr, 2453 struct uverbs_attr_bundle *attrs); 2454 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port, 2455 int state); 2456 int (*get_vf_config)(struct ib_device *device, int vf, u8 port, 2457 struct ifla_vf_info *ivf); 2458 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port, 2459 struct ifla_vf_stats *stats); 2460 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid, 2461 int type); 2462 struct ib_wq *(*create_wq)(struct ib_pd *pd, 2463 struct ib_wq_init_attr *init_attr, 2464 struct ib_udata *udata); 2465 int (*destroy_wq)(struct ib_wq *wq); 2466 int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr, 2467 u32 wq_attr_mask, struct ib_udata *udata); 2468 struct ib_rwq_ind_table *(*create_rwq_ind_table)( 2469 struct ib_device *device, 2470 struct ib_rwq_ind_table_init_attr *init_attr, 2471 struct ib_udata *udata); 2472 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table); 2473 struct ib_dm *(*alloc_dm)(struct ib_device *device, 2474 struct ib_ucontext *context, 2475 struct ib_dm_alloc_attr *attr, 2476 struct uverbs_attr_bundle *attrs); 2477 int (*dealloc_dm)(struct ib_dm *dm); 2478 struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm, 2479 struct ib_dm_mr_attr *attr, 2480 struct uverbs_attr_bundle *attrs); 2481 struct ib_counters *(*create_counters)( 2482 struct ib_device *device, struct uverbs_attr_bundle *attrs); 2483 int (*destroy_counters)(struct ib_counters *counters); 2484 int (*read_counters)(struct ib_counters *counters, 2485 struct ib_counters_read_attr *counters_read_attr, 2486 struct uverbs_attr_bundle *attrs); 2487 /** 2488 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the 2489 * driver initialized data. The struct is kfree()'ed by the sysfs 2490 * core when the device is removed. A lifespan of -1 in the return 2491 * struct tells the core to set a default lifespan. 2492 */ 2493 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device, 2494 u8 port_num); 2495 /** 2496 * get_hw_stats - Fill in the counter value(s) in the stats struct. 2497 * @index - The index in the value array we wish to have updated, or 2498 * num_counters if we want all stats updated 2499 * Return codes - 2500 * < 0 - Error, no counters updated 2501 * index - Updated the single counter pointed to by index 2502 * num_counters - Updated all counters (will reset the timestamp 2503 * and prevent further calls for lifespan milliseconds) 2504 * Drivers are allowed to update all counters in leiu of just the 2505 * one given in index at their option 2506 */ 2507 int (*get_hw_stats)(struct ib_device *device, 2508 struct rdma_hw_stats *stats, u8 port, int index); 2509 }; 2510 2511 struct ib_device { 2512 /* Do not access @dma_device directly from ULP nor from HW drivers. */ 2513 struct device *dma_device; 2514 struct ib_device_ops ops; 2515 char name[IB_DEVICE_NAME_MAX]; 2516 2517 struct list_head event_handler_list; 2518 spinlock_t event_handler_lock; 2519 2520 rwlock_t client_data_lock; 2521 struct list_head core_list; 2522 /* Access to the client_data_list is protected by the client_data_lock 2523 * rwlock and the lists_rwsem read-write semaphore 2524 */ 2525 struct list_head client_data_list; 2526 2527 struct ib_cache cache; 2528 /** 2529 * port_immutable is indexed by port number 2530 */ 2531 struct ib_port_immutable *port_immutable; 2532 2533 int num_comp_vectors; 2534 2535 struct ib_port_pkey_list *port_pkey_list; 2536 2537 struct iw_cm_verbs *iwcm; 2538 2539 struct module *owner; 2540 struct device dev; 2541 /* First group for device attributes, 2542 * Second group for driver provided attributes (optional). 2543 * It is NULL terminated array. 2544 */ 2545 const struct attribute_group *groups[3]; 2546 2547 struct kobject *ports_kobj; 2548 struct list_head port_list; 2549 2550 enum { 2551 IB_DEV_UNINITIALIZED, 2552 IB_DEV_REGISTERED, 2553 IB_DEV_UNREGISTERED 2554 } reg_state; 2555 2556 int uverbs_abi_ver; 2557 u64 uverbs_cmd_mask; 2558 u64 uverbs_ex_cmd_mask; 2559 2560 char node_desc[IB_DEVICE_NODE_DESC_MAX]; 2561 __be64 node_guid; 2562 u32 local_dma_lkey; 2563 u16 is_switch:1; 2564 u8 node_type; 2565 u8 phys_port_cnt; 2566 struct ib_device_attr attrs; 2567 struct attribute_group *hw_stats_ag; 2568 struct rdma_hw_stats *hw_stats; 2569 2570 #ifdef CONFIG_CGROUP_RDMA 2571 struct rdmacg_device cg_device; 2572 #endif 2573 2574 u32 index; 2575 /* 2576 * Implementation details of the RDMA core, don't use in drivers 2577 */ 2578 struct rdma_restrack_root res; 2579 2580 const struct uapi_definition *driver_def; 2581 enum rdma_driver_id driver_id; 2582 /* 2583 * Provides synchronization between device unregistration and netlink 2584 * commands on a device. To be used only by core. 2585 */ 2586 refcount_t refcount; 2587 struct completion unreg_completion; 2588 }; 2589 2590 struct ib_client { 2591 char *name; 2592 void (*add) (struct ib_device *); 2593 void (*remove)(struct ib_device *, void *client_data); 2594 2595 /* Returns the net_dev belonging to this ib_client and matching the 2596 * given parameters. 2597 * @dev: An RDMA device that the net_dev use for communication. 2598 * @port: A physical port number on the RDMA device. 2599 * @pkey: P_Key that the net_dev uses if applicable. 2600 * @gid: A GID that the net_dev uses to communicate. 2601 * @addr: An IP address the net_dev is configured with. 2602 * @client_data: The device's client data set by ib_set_client_data(). 2603 * 2604 * An ib_client that implements a net_dev on top of RDMA devices 2605 * (such as IP over IB) should implement this callback, allowing the 2606 * rdma_cm module to find the right net_dev for a given request. 2607 * 2608 * The caller is responsible for calling dev_put on the returned 2609 * netdev. */ 2610 struct net_device *(*get_net_dev_by_params)( 2611 struct ib_device *dev, 2612 u8 port, 2613 u16 pkey, 2614 const union ib_gid *gid, 2615 const struct sockaddr *addr, 2616 void *client_data); 2617 struct list_head list; 2618 }; 2619 2620 struct ib_device *ib_alloc_device(size_t size); 2621 void ib_dealloc_device(struct ib_device *device); 2622 2623 void ib_get_device_fw_str(struct ib_device *device, char *str); 2624 2625 int ib_register_device(struct ib_device *device, const char *name, 2626 int (*port_callback)(struct ib_device *, u8, 2627 struct kobject *)); 2628 void ib_unregister_device(struct ib_device *device); 2629 2630 int ib_register_client (struct ib_client *client); 2631 void ib_unregister_client(struct ib_client *client); 2632 2633 void *ib_get_client_data(struct ib_device *device, struct ib_client *client); 2634 void ib_set_client_data(struct ib_device *device, struct ib_client *client, 2635 void *data); 2636 void ib_set_device_ops(struct ib_device *device, 2637 const struct ib_device_ops *ops); 2638 2639 #if IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS) 2640 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma, 2641 unsigned long pfn, unsigned long size, pgprot_t prot); 2642 int rdma_user_mmap_page(struct ib_ucontext *ucontext, 2643 struct vm_area_struct *vma, struct page *page, 2644 unsigned long size); 2645 #else 2646 static inline int rdma_user_mmap_io(struct ib_ucontext *ucontext, 2647 struct vm_area_struct *vma, 2648 unsigned long pfn, unsigned long size, 2649 pgprot_t prot) 2650 { 2651 return -EINVAL; 2652 } 2653 static inline int rdma_user_mmap_page(struct ib_ucontext *ucontext, 2654 struct vm_area_struct *vma, struct page *page, 2655 unsigned long size) 2656 { 2657 return -EINVAL; 2658 } 2659 #endif 2660 2661 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len) 2662 { 2663 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0; 2664 } 2665 2666 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len) 2667 { 2668 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0; 2669 } 2670 2671 static inline bool ib_is_buffer_cleared(const void __user *p, 2672 size_t len) 2673 { 2674 bool ret; 2675 u8 *buf; 2676 2677 if (len > USHRT_MAX) 2678 return false; 2679 2680 buf = memdup_user(p, len); 2681 if (IS_ERR(buf)) 2682 return false; 2683 2684 ret = !memchr_inv(buf, 0, len); 2685 kfree(buf); 2686 return ret; 2687 } 2688 2689 static inline bool ib_is_udata_cleared(struct ib_udata *udata, 2690 size_t offset, 2691 size_t len) 2692 { 2693 return ib_is_buffer_cleared(udata->inbuf + offset, len); 2694 } 2695 2696 /** 2697 * ib_is_destroy_retryable - Check whether the uobject destruction 2698 * is retryable. 2699 * @ret: The initial destruction return code 2700 * @why: remove reason 2701 * @uobj: The uobject that is destroyed 2702 * 2703 * This function is a helper function that IB layer and low-level drivers 2704 * can use to consider whether the destruction of the given uobject is 2705 * retry-able. 2706 * It checks the original return code, if it wasn't success the destruction 2707 * is retryable according to the ucontext state (i.e. cleanup_retryable) and 2708 * the remove reason. (i.e. why). 2709 * Must be called with the object locked for destroy. 2710 */ 2711 static inline bool ib_is_destroy_retryable(int ret, enum rdma_remove_reason why, 2712 struct ib_uobject *uobj) 2713 { 2714 return ret && (why == RDMA_REMOVE_DESTROY || 2715 uobj->context->cleanup_retryable); 2716 } 2717 2718 /** 2719 * ib_destroy_usecnt - Called during destruction to check the usecnt 2720 * @usecnt: The usecnt atomic 2721 * @why: remove reason 2722 * @uobj: The uobject that is destroyed 2723 * 2724 * Non-zero usecnts will block destruction unless destruction was triggered by 2725 * a ucontext cleanup. 2726 */ 2727 static inline int ib_destroy_usecnt(atomic_t *usecnt, 2728 enum rdma_remove_reason why, 2729 struct ib_uobject *uobj) 2730 { 2731 if (atomic_read(usecnt) && ib_is_destroy_retryable(-EBUSY, why, uobj)) 2732 return -EBUSY; 2733 return 0; 2734 } 2735 2736 /** 2737 * ib_modify_qp_is_ok - Check that the supplied attribute mask 2738 * contains all required attributes and no attributes not allowed for 2739 * the given QP state transition. 2740 * @cur_state: Current QP state 2741 * @next_state: Next QP state 2742 * @type: QP type 2743 * @mask: Mask of supplied QP attributes 2744 * 2745 * This function is a helper function that a low-level driver's 2746 * modify_qp method can use to validate the consumer's input. It 2747 * checks that cur_state and next_state are valid QP states, that a 2748 * transition from cur_state to next_state is allowed by the IB spec, 2749 * and that the attribute mask supplied is allowed for the transition. 2750 */ 2751 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state, 2752 enum ib_qp_type type, enum ib_qp_attr_mask mask); 2753 2754 void ib_register_event_handler(struct ib_event_handler *event_handler); 2755 void ib_unregister_event_handler(struct ib_event_handler *event_handler); 2756 void ib_dispatch_event(struct ib_event *event); 2757 2758 int ib_query_port(struct ib_device *device, 2759 u8 port_num, struct ib_port_attr *port_attr); 2760 2761 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device, 2762 u8 port_num); 2763 2764 /** 2765 * rdma_cap_ib_switch - Check if the device is IB switch 2766 * @device: Device to check 2767 * 2768 * Device driver is responsible for setting is_switch bit on 2769 * in ib_device structure at init time. 2770 * 2771 * Return: true if the device is IB switch. 2772 */ 2773 static inline bool rdma_cap_ib_switch(const struct ib_device *device) 2774 { 2775 return device->is_switch; 2776 } 2777 2778 /** 2779 * rdma_start_port - Return the first valid port number for the device 2780 * specified 2781 * 2782 * @device: Device to be checked 2783 * 2784 * Return start port number 2785 */ 2786 static inline u8 rdma_start_port(const struct ib_device *device) 2787 { 2788 return rdma_cap_ib_switch(device) ? 0 : 1; 2789 } 2790 2791 /** 2792 * rdma_end_port - Return the last valid port number for the device 2793 * specified 2794 * 2795 * @device: Device to be checked 2796 * 2797 * Return last port number 2798 */ 2799 static inline u8 rdma_end_port(const struct ib_device *device) 2800 { 2801 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt; 2802 } 2803 2804 static inline int rdma_is_port_valid(const struct ib_device *device, 2805 unsigned int port) 2806 { 2807 return (port >= rdma_start_port(device) && 2808 port <= rdma_end_port(device)); 2809 } 2810 2811 static inline bool rdma_is_grh_required(const struct ib_device *device, 2812 u8 port_num) 2813 { 2814 return device->port_immutable[port_num].core_cap_flags & 2815 RDMA_CORE_PORT_IB_GRH_REQUIRED; 2816 } 2817 2818 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num) 2819 { 2820 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB; 2821 } 2822 2823 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num) 2824 { 2825 return device->port_immutable[port_num].core_cap_flags & 2826 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP); 2827 } 2828 2829 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num) 2830 { 2831 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP; 2832 } 2833 2834 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num) 2835 { 2836 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE; 2837 } 2838 2839 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num) 2840 { 2841 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP; 2842 } 2843 2844 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num) 2845 { 2846 return rdma_protocol_ib(device, port_num) || 2847 rdma_protocol_roce(device, port_num); 2848 } 2849 2850 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num) 2851 { 2852 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_RAW_PACKET; 2853 } 2854 2855 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num) 2856 { 2857 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_USNIC; 2858 } 2859 2860 /** 2861 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband 2862 * Management Datagrams. 2863 * @device: Device to check 2864 * @port_num: Port number to check 2865 * 2866 * Management Datagrams (MAD) are a required part of the InfiniBand 2867 * specification and are supported on all InfiniBand devices. A slightly 2868 * extended version are also supported on OPA interfaces. 2869 * 2870 * Return: true if the port supports sending/receiving of MAD packets. 2871 */ 2872 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num) 2873 { 2874 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD; 2875 } 2876 2877 /** 2878 * rdma_cap_opa_mad - Check if the port of device provides support for OPA 2879 * Management Datagrams. 2880 * @device: Device to check 2881 * @port_num: Port number to check 2882 * 2883 * Intel OmniPath devices extend and/or replace the InfiniBand Management 2884 * datagrams with their own versions. These OPA MADs share many but not all of 2885 * the characteristics of InfiniBand MADs. 2886 * 2887 * OPA MADs differ in the following ways: 2888 * 2889 * 1) MADs are variable size up to 2K 2890 * IBTA defined MADs remain fixed at 256 bytes 2891 * 2) OPA SMPs must carry valid PKeys 2892 * 3) OPA SMP packets are a different format 2893 * 2894 * Return: true if the port supports OPA MAD packet formats. 2895 */ 2896 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num) 2897 { 2898 return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD) 2899 == RDMA_CORE_CAP_OPA_MAD; 2900 } 2901 2902 /** 2903 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband 2904 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI). 2905 * @device: Device to check 2906 * @port_num: Port number to check 2907 * 2908 * Each InfiniBand node is required to provide a Subnet Management Agent 2909 * that the subnet manager can access. Prior to the fabric being fully 2910 * configured by the subnet manager, the SMA is accessed via a well known 2911 * interface called the Subnet Management Interface (SMI). This interface 2912 * uses directed route packets to communicate with the SM to get around the 2913 * chicken and egg problem of the SM needing to know what's on the fabric 2914 * in order to configure the fabric, and needing to configure the fabric in 2915 * order to send packets to the devices on the fabric. These directed 2916 * route packets do not need the fabric fully configured in order to reach 2917 * their destination. The SMI is the only method allowed to send 2918 * directed route packets on an InfiniBand fabric. 2919 * 2920 * Return: true if the port provides an SMI. 2921 */ 2922 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num) 2923 { 2924 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI; 2925 } 2926 2927 /** 2928 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband 2929 * Communication Manager. 2930 * @device: Device to check 2931 * @port_num: Port number to check 2932 * 2933 * The InfiniBand Communication Manager is one of many pre-defined General 2934 * Service Agents (GSA) that are accessed via the General Service 2935 * Interface (GSI). It's role is to facilitate establishment of connections 2936 * between nodes as well as other management related tasks for established 2937 * connections. 2938 * 2939 * Return: true if the port supports an IB CM (this does not guarantee that 2940 * a CM is actually running however). 2941 */ 2942 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num) 2943 { 2944 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM; 2945 } 2946 2947 /** 2948 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP 2949 * Communication Manager. 2950 * @device: Device to check 2951 * @port_num: Port number to check 2952 * 2953 * Similar to above, but specific to iWARP connections which have a different 2954 * managment protocol than InfiniBand. 2955 * 2956 * Return: true if the port supports an iWARP CM (this does not guarantee that 2957 * a CM is actually running however). 2958 */ 2959 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num) 2960 { 2961 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM; 2962 } 2963 2964 /** 2965 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband 2966 * Subnet Administration. 2967 * @device: Device to check 2968 * @port_num: Port number to check 2969 * 2970 * An InfiniBand Subnet Administration (SA) service is a pre-defined General 2971 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand 2972 * fabrics, devices should resolve routes to other hosts by contacting the 2973 * SA to query the proper route. 2974 * 2975 * Return: true if the port should act as a client to the fabric Subnet 2976 * Administration interface. This does not imply that the SA service is 2977 * running locally. 2978 */ 2979 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num) 2980 { 2981 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA; 2982 } 2983 2984 /** 2985 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband 2986 * Multicast. 2987 * @device: Device to check 2988 * @port_num: Port number to check 2989 * 2990 * InfiniBand multicast registration is more complex than normal IPv4 or 2991 * IPv6 multicast registration. Each Host Channel Adapter must register 2992 * with the Subnet Manager when it wishes to join a multicast group. It 2993 * should do so only once regardless of how many queue pairs it subscribes 2994 * to this group. And it should leave the group only after all queue pairs 2995 * attached to the group have been detached. 2996 * 2997 * Return: true if the port must undertake the additional adminstrative 2998 * overhead of registering/unregistering with the SM and tracking of the 2999 * total number of queue pairs attached to the multicast group. 3000 */ 3001 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num) 3002 { 3003 return rdma_cap_ib_sa(device, port_num); 3004 } 3005 3006 /** 3007 * rdma_cap_af_ib - Check if the port of device has the capability 3008 * Native Infiniband Address. 3009 * @device: Device to check 3010 * @port_num: Port number to check 3011 * 3012 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default 3013 * GID. RoCE uses a different mechanism, but still generates a GID via 3014 * a prescribed mechanism and port specific data. 3015 * 3016 * Return: true if the port uses a GID address to identify devices on the 3017 * network. 3018 */ 3019 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num) 3020 { 3021 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB; 3022 } 3023 3024 /** 3025 * rdma_cap_eth_ah - Check if the port of device has the capability 3026 * Ethernet Address Handle. 3027 * @device: Device to check 3028 * @port_num: Port number to check 3029 * 3030 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique 3031 * to fabricate GIDs over Ethernet/IP specific addresses native to the 3032 * port. Normally, packet headers are generated by the sending host 3033 * adapter, but when sending connectionless datagrams, we must manually 3034 * inject the proper headers for the fabric we are communicating over. 3035 * 3036 * Return: true if we are running as a RoCE port and must force the 3037 * addition of a Global Route Header built from our Ethernet Address 3038 * Handle into our header list for connectionless packets. 3039 */ 3040 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num) 3041 { 3042 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH; 3043 } 3044 3045 /** 3046 * rdma_cap_opa_ah - Check if the port of device supports 3047 * OPA Address handles 3048 * @device: Device to check 3049 * @port_num: Port number to check 3050 * 3051 * Return: true if we are running on an OPA device which supports 3052 * the extended OPA addressing. 3053 */ 3054 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num) 3055 { 3056 return (device->port_immutable[port_num].core_cap_flags & 3057 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH; 3058 } 3059 3060 /** 3061 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port. 3062 * 3063 * @device: Device 3064 * @port_num: Port number 3065 * 3066 * This MAD size includes the MAD headers and MAD payload. No other headers 3067 * are included. 3068 * 3069 * Return the max MAD size required by the Port. Will return 0 if the port 3070 * does not support MADs 3071 */ 3072 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num) 3073 { 3074 return device->port_immutable[port_num].max_mad_size; 3075 } 3076 3077 /** 3078 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table 3079 * @device: Device to check 3080 * @port_num: Port number to check 3081 * 3082 * RoCE GID table mechanism manages the various GIDs for a device. 3083 * 3084 * NOTE: if allocating the port's GID table has failed, this call will still 3085 * return true, but any RoCE GID table API will fail. 3086 * 3087 * Return: true if the port uses RoCE GID table mechanism in order to manage 3088 * its GIDs. 3089 */ 3090 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device, 3091 u8 port_num) 3092 { 3093 return rdma_protocol_roce(device, port_num) && 3094 device->ops.add_gid && device->ops.del_gid; 3095 } 3096 3097 /* 3098 * Check if the device supports READ W/ INVALIDATE. 3099 */ 3100 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num) 3101 { 3102 /* 3103 * iWarp drivers must support READ W/ INVALIDATE. No other protocol 3104 * has support for it yet. 3105 */ 3106 return rdma_protocol_iwarp(dev, port_num); 3107 } 3108 3109 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port, 3110 int state); 3111 int ib_get_vf_config(struct ib_device *device, int vf, u8 port, 3112 struct ifla_vf_info *info); 3113 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port, 3114 struct ifla_vf_stats *stats); 3115 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid, 3116 int type); 3117 3118 int ib_query_pkey(struct ib_device *device, 3119 u8 port_num, u16 index, u16 *pkey); 3120 3121 int ib_modify_device(struct ib_device *device, 3122 int device_modify_mask, 3123 struct ib_device_modify *device_modify); 3124 3125 int ib_modify_port(struct ib_device *device, 3126 u8 port_num, int port_modify_mask, 3127 struct ib_port_modify *port_modify); 3128 3129 int ib_find_gid(struct ib_device *device, union ib_gid *gid, 3130 u8 *port_num, u16 *index); 3131 3132 int ib_find_pkey(struct ib_device *device, 3133 u8 port_num, u16 pkey, u16 *index); 3134 3135 enum ib_pd_flags { 3136 /* 3137 * Create a memory registration for all memory in the system and place 3138 * the rkey for it into pd->unsafe_global_rkey. This can be used by 3139 * ULPs to avoid the overhead of dynamic MRs. 3140 * 3141 * This flag is generally considered unsafe and must only be used in 3142 * extremly trusted environments. Every use of it will log a warning 3143 * in the kernel log. 3144 */ 3145 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01, 3146 }; 3147 3148 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags, 3149 const char *caller); 3150 #define ib_alloc_pd(device, flags) \ 3151 __ib_alloc_pd((device), (flags), KBUILD_MODNAME) 3152 void ib_dealloc_pd(struct ib_pd *pd); 3153 3154 enum rdma_create_ah_flags { 3155 /* In a sleepable context */ 3156 RDMA_CREATE_AH_SLEEPABLE = BIT(0), 3157 }; 3158 3159 /** 3160 * rdma_create_ah - Creates an address handle for the given address vector. 3161 * @pd: The protection domain associated with the address handle. 3162 * @ah_attr: The attributes of the address vector. 3163 * @flags: Create address handle flags (see enum rdma_create_ah_flags). 3164 * 3165 * The address handle is used to reference a local or global destination 3166 * in all UD QP post sends. 3167 */ 3168 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr, 3169 u32 flags); 3170 3171 /** 3172 * rdma_create_user_ah - Creates an address handle for the given address vector. 3173 * It resolves destination mac address for ah attribute of RoCE type. 3174 * @pd: The protection domain associated with the address handle. 3175 * @ah_attr: The attributes of the address vector. 3176 * @udata: pointer to user's input output buffer information need by 3177 * provider driver. 3178 * 3179 * It returns 0 on success and returns appropriate error code on error. 3180 * The address handle is used to reference a local or global destination 3181 * in all UD QP post sends. 3182 */ 3183 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd, 3184 struct rdma_ah_attr *ah_attr, 3185 struct ib_udata *udata); 3186 /** 3187 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header 3188 * work completion. 3189 * @hdr: the L3 header to parse 3190 * @net_type: type of header to parse 3191 * @sgid: place to store source gid 3192 * @dgid: place to store destination gid 3193 */ 3194 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr, 3195 enum rdma_network_type net_type, 3196 union ib_gid *sgid, union ib_gid *dgid); 3197 3198 /** 3199 * ib_get_rdma_header_version - Get the header version 3200 * @hdr: the L3 header to parse 3201 */ 3202 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr); 3203 3204 /** 3205 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a 3206 * work completion. 3207 * @device: Device on which the received message arrived. 3208 * @port_num: Port on which the received message arrived. 3209 * @wc: Work completion associated with the received message. 3210 * @grh: References the received global route header. This parameter is 3211 * ignored unless the work completion indicates that the GRH is valid. 3212 * @ah_attr: Returned attributes that can be used when creating an address 3213 * handle for replying to the message. 3214 * When ib_init_ah_attr_from_wc() returns success, 3215 * (a) for IB link layer it optionally contains a reference to SGID attribute 3216 * when GRH is present for IB link layer. 3217 * (b) for RoCE link layer it contains a reference to SGID attribute. 3218 * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID 3219 * attributes which are initialized using ib_init_ah_attr_from_wc(). 3220 * 3221 */ 3222 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num, 3223 const struct ib_wc *wc, const struct ib_grh *grh, 3224 struct rdma_ah_attr *ah_attr); 3225 3226 /** 3227 * ib_create_ah_from_wc - Creates an address handle associated with the 3228 * sender of the specified work completion. 3229 * @pd: The protection domain associated with the address handle. 3230 * @wc: Work completion information associated with a received message. 3231 * @grh: References the received global route header. This parameter is 3232 * ignored unless the work completion indicates that the GRH is valid. 3233 * @port_num: The outbound port number to associate with the address. 3234 * 3235 * The address handle is used to reference a local or global destination 3236 * in all UD QP post sends. 3237 */ 3238 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc, 3239 const struct ib_grh *grh, u8 port_num); 3240 3241 /** 3242 * rdma_modify_ah - Modifies the address vector associated with an address 3243 * handle. 3244 * @ah: The address handle to modify. 3245 * @ah_attr: The new address vector attributes to associate with the 3246 * address handle. 3247 */ 3248 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr); 3249 3250 /** 3251 * rdma_query_ah - Queries the address vector associated with an address 3252 * handle. 3253 * @ah: The address handle to query. 3254 * @ah_attr: The address vector attributes associated with the address 3255 * handle. 3256 */ 3257 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr); 3258 3259 enum rdma_destroy_ah_flags { 3260 /* In a sleepable context */ 3261 RDMA_DESTROY_AH_SLEEPABLE = BIT(0), 3262 }; 3263 3264 /** 3265 * rdma_destroy_ah - Destroys an address handle. 3266 * @ah: The address handle to destroy. 3267 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags). 3268 */ 3269 int rdma_destroy_ah(struct ib_ah *ah, u32 flags); 3270 3271 /** 3272 * ib_create_srq - Creates a SRQ associated with the specified protection 3273 * domain. 3274 * @pd: The protection domain associated with the SRQ. 3275 * @srq_init_attr: A list of initial attributes required to create the 3276 * SRQ. If SRQ creation succeeds, then the attributes are updated to 3277 * the actual capabilities of the created SRQ. 3278 * 3279 * srq_attr->max_wr and srq_attr->max_sge are read the determine the 3280 * requested size of the SRQ, and set to the actual values allocated 3281 * on return. If ib_create_srq() succeeds, then max_wr and max_sge 3282 * will always be at least as large as the requested values. 3283 */ 3284 struct ib_srq *ib_create_srq(struct ib_pd *pd, 3285 struct ib_srq_init_attr *srq_init_attr); 3286 3287 /** 3288 * ib_modify_srq - Modifies the attributes for the specified SRQ. 3289 * @srq: The SRQ to modify. 3290 * @srq_attr: On input, specifies the SRQ attributes to modify. On output, 3291 * the current values of selected SRQ attributes are returned. 3292 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ 3293 * are being modified. 3294 * 3295 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or 3296 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when 3297 * the number of receives queued drops below the limit. 3298 */ 3299 int ib_modify_srq(struct ib_srq *srq, 3300 struct ib_srq_attr *srq_attr, 3301 enum ib_srq_attr_mask srq_attr_mask); 3302 3303 /** 3304 * ib_query_srq - Returns the attribute list and current values for the 3305 * specified SRQ. 3306 * @srq: The SRQ to query. 3307 * @srq_attr: The attributes of the specified SRQ. 3308 */ 3309 int ib_query_srq(struct ib_srq *srq, 3310 struct ib_srq_attr *srq_attr); 3311 3312 /** 3313 * ib_destroy_srq - Destroys the specified SRQ. 3314 * @srq: The SRQ to destroy. 3315 */ 3316 int ib_destroy_srq(struct ib_srq *srq); 3317 3318 /** 3319 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ. 3320 * @srq: The SRQ to post the work request on. 3321 * @recv_wr: A list of work requests to post on the receive queue. 3322 * @bad_recv_wr: On an immediate failure, this parameter will reference 3323 * the work request that failed to be posted on the QP. 3324 */ 3325 static inline int ib_post_srq_recv(struct ib_srq *srq, 3326 const struct ib_recv_wr *recv_wr, 3327 const struct ib_recv_wr **bad_recv_wr) 3328 { 3329 const struct ib_recv_wr *dummy; 3330 3331 return srq->device->ops.post_srq_recv(srq, recv_wr, 3332 bad_recv_wr ? : &dummy); 3333 } 3334 3335 /** 3336 * ib_create_qp - Creates a QP associated with the specified protection 3337 * domain. 3338 * @pd: The protection domain associated with the QP. 3339 * @qp_init_attr: A list of initial attributes required to create the 3340 * QP. If QP creation succeeds, then the attributes are updated to 3341 * the actual capabilities of the created QP. 3342 */ 3343 struct ib_qp *ib_create_qp(struct ib_pd *pd, 3344 struct ib_qp_init_attr *qp_init_attr); 3345 3346 /** 3347 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP. 3348 * @qp: The QP to modify. 3349 * @attr: On input, specifies the QP attributes to modify. On output, 3350 * the current values of selected QP attributes are returned. 3351 * @attr_mask: A bit-mask used to specify which attributes of the QP 3352 * are being modified. 3353 * @udata: pointer to user's input output buffer information 3354 * are being modified. 3355 * It returns 0 on success and returns appropriate error code on error. 3356 */ 3357 int ib_modify_qp_with_udata(struct ib_qp *qp, 3358 struct ib_qp_attr *attr, 3359 int attr_mask, 3360 struct ib_udata *udata); 3361 3362 /** 3363 * ib_modify_qp - Modifies the attributes for the specified QP and then 3364 * transitions the QP to the given state. 3365 * @qp: The QP to modify. 3366 * @qp_attr: On input, specifies the QP attributes to modify. On output, 3367 * the current values of selected QP attributes are returned. 3368 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP 3369 * are being modified. 3370 */ 3371 int ib_modify_qp(struct ib_qp *qp, 3372 struct ib_qp_attr *qp_attr, 3373 int qp_attr_mask); 3374 3375 /** 3376 * ib_query_qp - Returns the attribute list and current values for the 3377 * specified QP. 3378 * @qp: The QP to query. 3379 * @qp_attr: The attributes of the specified QP. 3380 * @qp_attr_mask: A bit-mask used to select specific attributes to query. 3381 * @qp_init_attr: Additional attributes of the selected QP. 3382 * 3383 * The qp_attr_mask may be used to limit the query to gathering only the 3384 * selected attributes. 3385 */ 3386 int ib_query_qp(struct ib_qp *qp, 3387 struct ib_qp_attr *qp_attr, 3388 int qp_attr_mask, 3389 struct ib_qp_init_attr *qp_init_attr); 3390 3391 /** 3392 * ib_destroy_qp - Destroys the specified QP. 3393 * @qp: The QP to destroy. 3394 */ 3395 int ib_destroy_qp(struct ib_qp *qp); 3396 3397 /** 3398 * ib_open_qp - Obtain a reference to an existing sharable QP. 3399 * @xrcd - XRC domain 3400 * @qp_open_attr: Attributes identifying the QP to open. 3401 * 3402 * Returns a reference to a sharable QP. 3403 */ 3404 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd, 3405 struct ib_qp_open_attr *qp_open_attr); 3406 3407 /** 3408 * ib_close_qp - Release an external reference to a QP. 3409 * @qp: The QP handle to release 3410 * 3411 * The opened QP handle is released by the caller. The underlying 3412 * shared QP is not destroyed until all internal references are released. 3413 */ 3414 int ib_close_qp(struct ib_qp *qp); 3415 3416 /** 3417 * ib_post_send - Posts a list of work requests to the send queue of 3418 * the specified QP. 3419 * @qp: The QP to post the work request on. 3420 * @send_wr: A list of work requests to post on the send queue. 3421 * @bad_send_wr: On an immediate failure, this parameter will reference 3422 * the work request that failed to be posted on the QP. 3423 * 3424 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate 3425 * error is returned, the QP state shall not be affected, 3426 * ib_post_send() will return an immediate error after queueing any 3427 * earlier work requests in the list. 3428 */ 3429 static inline int ib_post_send(struct ib_qp *qp, 3430 const struct ib_send_wr *send_wr, 3431 const struct ib_send_wr **bad_send_wr) 3432 { 3433 const struct ib_send_wr *dummy; 3434 3435 return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy); 3436 } 3437 3438 /** 3439 * ib_post_recv - Posts a list of work requests to the receive queue of 3440 * the specified QP. 3441 * @qp: The QP to post the work request on. 3442 * @recv_wr: A list of work requests to post on the receive queue. 3443 * @bad_recv_wr: On an immediate failure, this parameter will reference 3444 * the work request that failed to be posted on the QP. 3445 */ 3446 static inline int ib_post_recv(struct ib_qp *qp, 3447 const struct ib_recv_wr *recv_wr, 3448 const struct ib_recv_wr **bad_recv_wr) 3449 { 3450 const struct ib_recv_wr *dummy; 3451 3452 return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy); 3453 } 3454 3455 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private, 3456 int nr_cqe, int comp_vector, 3457 enum ib_poll_context poll_ctx, const char *caller); 3458 #define ib_alloc_cq(device, priv, nr_cqe, comp_vect, poll_ctx) \ 3459 __ib_alloc_cq((device), (priv), (nr_cqe), (comp_vect), (poll_ctx), KBUILD_MODNAME) 3460 3461 void ib_free_cq(struct ib_cq *cq); 3462 int ib_process_cq_direct(struct ib_cq *cq, int budget); 3463 3464 /** 3465 * ib_create_cq - Creates a CQ on the specified device. 3466 * @device: The device on which to create the CQ. 3467 * @comp_handler: A user-specified callback that is invoked when a 3468 * completion event occurs on the CQ. 3469 * @event_handler: A user-specified callback that is invoked when an 3470 * asynchronous event not associated with a completion occurs on the CQ. 3471 * @cq_context: Context associated with the CQ returned to the user via 3472 * the associated completion and event handlers. 3473 * @cq_attr: The attributes the CQ should be created upon. 3474 * 3475 * Users can examine the cq structure to determine the actual CQ size. 3476 */ 3477 struct ib_cq *__ib_create_cq(struct ib_device *device, 3478 ib_comp_handler comp_handler, 3479 void (*event_handler)(struct ib_event *, void *), 3480 void *cq_context, 3481 const struct ib_cq_init_attr *cq_attr, 3482 const char *caller); 3483 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \ 3484 __ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME) 3485 3486 /** 3487 * ib_resize_cq - Modifies the capacity of the CQ. 3488 * @cq: The CQ to resize. 3489 * @cqe: The minimum size of the CQ. 3490 * 3491 * Users can examine the cq structure to determine the actual CQ size. 3492 */ 3493 int ib_resize_cq(struct ib_cq *cq, int cqe); 3494 3495 /** 3496 * rdma_set_cq_moderation - Modifies moderation params of the CQ 3497 * @cq: The CQ to modify. 3498 * @cq_count: number of CQEs that will trigger an event 3499 * @cq_period: max period of time in usec before triggering an event 3500 * 3501 */ 3502 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period); 3503 3504 /** 3505 * ib_destroy_cq - Destroys the specified CQ. 3506 * @cq: The CQ to destroy. 3507 */ 3508 int ib_destroy_cq(struct ib_cq *cq); 3509 3510 /** 3511 * ib_poll_cq - poll a CQ for completion(s) 3512 * @cq:the CQ being polled 3513 * @num_entries:maximum number of completions to return 3514 * @wc:array of at least @num_entries &struct ib_wc where completions 3515 * will be returned 3516 * 3517 * Poll a CQ for (possibly multiple) completions. If the return value 3518 * is < 0, an error occurred. If the return value is >= 0, it is the 3519 * number of completions returned. If the return value is 3520 * non-negative and < num_entries, then the CQ was emptied. 3521 */ 3522 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries, 3523 struct ib_wc *wc) 3524 { 3525 return cq->device->ops.poll_cq(cq, num_entries, wc); 3526 } 3527 3528 /** 3529 * ib_req_notify_cq - Request completion notification on a CQ. 3530 * @cq: The CQ to generate an event for. 3531 * @flags: 3532 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP 3533 * to request an event on the next solicited event or next work 3534 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS 3535 * may also be |ed in to request a hint about missed events, as 3536 * described below. 3537 * 3538 * Return Value: 3539 * < 0 means an error occurred while requesting notification 3540 * == 0 means notification was requested successfully, and if 3541 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events 3542 * were missed and it is safe to wait for another event. In 3543 * this case is it guaranteed that any work completions added 3544 * to the CQ since the last CQ poll will trigger a completion 3545 * notification event. 3546 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed 3547 * in. It means that the consumer must poll the CQ again to 3548 * make sure it is empty to avoid missing an event because of a 3549 * race between requesting notification and an entry being 3550 * added to the CQ. This return value means it is possible 3551 * (but not guaranteed) that a work completion has been added 3552 * to the CQ since the last poll without triggering a 3553 * completion notification event. 3554 */ 3555 static inline int ib_req_notify_cq(struct ib_cq *cq, 3556 enum ib_cq_notify_flags flags) 3557 { 3558 return cq->device->ops.req_notify_cq(cq, flags); 3559 } 3560 3561 /** 3562 * ib_req_ncomp_notif - Request completion notification when there are 3563 * at least the specified number of unreaped completions on the CQ. 3564 * @cq: The CQ to generate an event for. 3565 * @wc_cnt: The number of unreaped completions that should be on the 3566 * CQ before an event is generated. 3567 */ 3568 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt) 3569 { 3570 return cq->device->ops.req_ncomp_notif ? 3571 cq->device->ops.req_ncomp_notif(cq, wc_cnt) : 3572 -ENOSYS; 3573 } 3574 3575 /** 3576 * ib_dma_mapping_error - check a DMA addr for error 3577 * @dev: The device for which the dma_addr was created 3578 * @dma_addr: The DMA address to check 3579 */ 3580 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr) 3581 { 3582 return dma_mapping_error(dev->dma_device, dma_addr); 3583 } 3584 3585 /** 3586 * ib_dma_map_single - Map a kernel virtual address to DMA address 3587 * @dev: The device for which the dma_addr is to be created 3588 * @cpu_addr: The kernel virtual address 3589 * @size: The size of the region in bytes 3590 * @direction: The direction of the DMA 3591 */ 3592 static inline u64 ib_dma_map_single(struct ib_device *dev, 3593 void *cpu_addr, size_t size, 3594 enum dma_data_direction direction) 3595 { 3596 return dma_map_single(dev->dma_device, cpu_addr, size, direction); 3597 } 3598 3599 /** 3600 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single() 3601 * @dev: The device for which the DMA address was created 3602 * @addr: The DMA address 3603 * @size: The size of the region in bytes 3604 * @direction: The direction of the DMA 3605 */ 3606 static inline void ib_dma_unmap_single(struct ib_device *dev, 3607 u64 addr, size_t size, 3608 enum dma_data_direction direction) 3609 { 3610 dma_unmap_single(dev->dma_device, addr, size, direction); 3611 } 3612 3613 /** 3614 * ib_dma_map_page - Map a physical page to DMA address 3615 * @dev: The device for which the dma_addr is to be created 3616 * @page: The page to be mapped 3617 * @offset: The offset within the page 3618 * @size: The size of the region in bytes 3619 * @direction: The direction of the DMA 3620 */ 3621 static inline u64 ib_dma_map_page(struct ib_device *dev, 3622 struct page *page, 3623 unsigned long offset, 3624 size_t size, 3625 enum dma_data_direction direction) 3626 { 3627 return dma_map_page(dev->dma_device, page, offset, size, direction); 3628 } 3629 3630 /** 3631 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page() 3632 * @dev: The device for which the DMA address was created 3633 * @addr: The DMA address 3634 * @size: The size of the region in bytes 3635 * @direction: The direction of the DMA 3636 */ 3637 static inline void ib_dma_unmap_page(struct ib_device *dev, 3638 u64 addr, size_t size, 3639 enum dma_data_direction direction) 3640 { 3641 dma_unmap_page(dev->dma_device, addr, size, direction); 3642 } 3643 3644 /** 3645 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses 3646 * @dev: The device for which the DMA addresses are to be created 3647 * @sg: The array of scatter/gather entries 3648 * @nents: The number of scatter/gather entries 3649 * @direction: The direction of the DMA 3650 */ 3651 static inline int ib_dma_map_sg(struct ib_device *dev, 3652 struct scatterlist *sg, int nents, 3653 enum dma_data_direction direction) 3654 { 3655 return dma_map_sg(dev->dma_device, sg, nents, direction); 3656 } 3657 3658 /** 3659 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses 3660 * @dev: The device for which the DMA addresses were created 3661 * @sg: The array of scatter/gather entries 3662 * @nents: The number of scatter/gather entries 3663 * @direction: The direction of the DMA 3664 */ 3665 static inline void ib_dma_unmap_sg(struct ib_device *dev, 3666 struct scatterlist *sg, int nents, 3667 enum dma_data_direction direction) 3668 { 3669 dma_unmap_sg(dev->dma_device, sg, nents, direction); 3670 } 3671 3672 static inline int ib_dma_map_sg_attrs(struct ib_device *dev, 3673 struct scatterlist *sg, int nents, 3674 enum dma_data_direction direction, 3675 unsigned long dma_attrs) 3676 { 3677 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, 3678 dma_attrs); 3679 } 3680 3681 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev, 3682 struct scatterlist *sg, int nents, 3683 enum dma_data_direction direction, 3684 unsigned long dma_attrs) 3685 { 3686 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs); 3687 } 3688 /** 3689 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry 3690 * @dev: The device for which the DMA addresses were created 3691 * @sg: The scatter/gather entry 3692 * 3693 * Note: this function is obsolete. To do: change all occurrences of 3694 * ib_sg_dma_address() into sg_dma_address(). 3695 */ 3696 static inline u64 ib_sg_dma_address(struct ib_device *dev, 3697 struct scatterlist *sg) 3698 { 3699 return sg_dma_address(sg); 3700 } 3701 3702 /** 3703 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry 3704 * @dev: The device for which the DMA addresses were created 3705 * @sg: The scatter/gather entry 3706 * 3707 * Note: this function is obsolete. To do: change all occurrences of 3708 * ib_sg_dma_len() into sg_dma_len(). 3709 */ 3710 static inline unsigned int ib_sg_dma_len(struct ib_device *dev, 3711 struct scatterlist *sg) 3712 { 3713 return sg_dma_len(sg); 3714 } 3715 3716 /** 3717 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU 3718 * @dev: The device for which the DMA address was created 3719 * @addr: The DMA address 3720 * @size: The size of the region in bytes 3721 * @dir: The direction of the DMA 3722 */ 3723 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev, 3724 u64 addr, 3725 size_t size, 3726 enum dma_data_direction dir) 3727 { 3728 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir); 3729 } 3730 3731 /** 3732 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device 3733 * @dev: The device for which the DMA address was created 3734 * @addr: The DMA address 3735 * @size: The size of the region in bytes 3736 * @dir: The direction of the DMA 3737 */ 3738 static inline void ib_dma_sync_single_for_device(struct ib_device *dev, 3739 u64 addr, 3740 size_t size, 3741 enum dma_data_direction dir) 3742 { 3743 dma_sync_single_for_device(dev->dma_device, addr, size, dir); 3744 } 3745 3746 /** 3747 * ib_dma_alloc_coherent - Allocate memory and map it for DMA 3748 * @dev: The device for which the DMA address is requested 3749 * @size: The size of the region to allocate in bytes 3750 * @dma_handle: A pointer for returning the DMA address of the region 3751 * @flag: memory allocator flags 3752 */ 3753 static inline void *ib_dma_alloc_coherent(struct ib_device *dev, 3754 size_t size, 3755 dma_addr_t *dma_handle, 3756 gfp_t flag) 3757 { 3758 return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag); 3759 } 3760 3761 /** 3762 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent() 3763 * @dev: The device for which the DMA addresses were allocated 3764 * @size: The size of the region 3765 * @cpu_addr: the address returned by ib_dma_alloc_coherent() 3766 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent() 3767 */ 3768 static inline void ib_dma_free_coherent(struct ib_device *dev, 3769 size_t size, void *cpu_addr, 3770 dma_addr_t dma_handle) 3771 { 3772 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle); 3773 } 3774 3775 /** 3776 * ib_dereg_mr - Deregisters a memory region and removes it from the 3777 * HCA translation table. 3778 * @mr: The memory region to deregister. 3779 * 3780 * This function can fail, if the memory region has memory windows bound to it. 3781 */ 3782 int ib_dereg_mr(struct ib_mr *mr); 3783 3784 struct ib_mr *ib_alloc_mr(struct ib_pd *pd, 3785 enum ib_mr_type mr_type, 3786 u32 max_num_sg); 3787 3788 /** 3789 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR 3790 * R_Key and L_Key. 3791 * @mr - struct ib_mr pointer to be updated. 3792 * @newkey - new key to be used. 3793 */ 3794 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey) 3795 { 3796 mr->lkey = (mr->lkey & 0xffffff00) | newkey; 3797 mr->rkey = (mr->rkey & 0xffffff00) | newkey; 3798 } 3799 3800 /** 3801 * ib_inc_rkey - increments the key portion of the given rkey. Can be used 3802 * for calculating a new rkey for type 2 memory windows. 3803 * @rkey - the rkey to increment. 3804 */ 3805 static inline u32 ib_inc_rkey(u32 rkey) 3806 { 3807 const u32 mask = 0x000000ff; 3808 return ((rkey + 1) & mask) | (rkey & ~mask); 3809 } 3810 3811 /** 3812 * ib_alloc_fmr - Allocates a unmapped fast memory region. 3813 * @pd: The protection domain associated with the unmapped region. 3814 * @mr_access_flags: Specifies the memory access rights. 3815 * @fmr_attr: Attributes of the unmapped region. 3816 * 3817 * A fast memory region must be mapped before it can be used as part of 3818 * a work request. 3819 */ 3820 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd, 3821 int mr_access_flags, 3822 struct ib_fmr_attr *fmr_attr); 3823 3824 /** 3825 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region. 3826 * @fmr: The fast memory region to associate with the pages. 3827 * @page_list: An array of physical pages to map to the fast memory region. 3828 * @list_len: The number of pages in page_list. 3829 * @iova: The I/O virtual address to use with the mapped region. 3830 */ 3831 static inline int ib_map_phys_fmr(struct ib_fmr *fmr, 3832 u64 *page_list, int list_len, 3833 u64 iova) 3834 { 3835 return fmr->device->ops.map_phys_fmr(fmr, page_list, list_len, iova); 3836 } 3837 3838 /** 3839 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions. 3840 * @fmr_list: A linked list of fast memory regions to unmap. 3841 */ 3842 int ib_unmap_fmr(struct list_head *fmr_list); 3843 3844 /** 3845 * ib_dealloc_fmr - Deallocates a fast memory region. 3846 * @fmr: The fast memory region to deallocate. 3847 */ 3848 int ib_dealloc_fmr(struct ib_fmr *fmr); 3849 3850 /** 3851 * ib_attach_mcast - Attaches the specified QP to a multicast group. 3852 * @qp: QP to attach to the multicast group. The QP must be type 3853 * IB_QPT_UD. 3854 * @gid: Multicast group GID. 3855 * @lid: Multicast group LID in host byte order. 3856 * 3857 * In order to send and receive multicast packets, subnet 3858 * administration must have created the multicast group and configured 3859 * the fabric appropriately. The port associated with the specified 3860 * QP must also be a member of the multicast group. 3861 */ 3862 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid); 3863 3864 /** 3865 * ib_detach_mcast - Detaches the specified QP from a multicast group. 3866 * @qp: QP to detach from the multicast group. 3867 * @gid: Multicast group GID. 3868 * @lid: Multicast group LID in host byte order. 3869 */ 3870 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid); 3871 3872 /** 3873 * ib_alloc_xrcd - Allocates an XRC domain. 3874 * @device: The device on which to allocate the XRC domain. 3875 * @caller: Module name for kernel consumers 3876 */ 3877 struct ib_xrcd *__ib_alloc_xrcd(struct ib_device *device, const char *caller); 3878 #define ib_alloc_xrcd(device) \ 3879 __ib_alloc_xrcd((device), KBUILD_MODNAME) 3880 3881 /** 3882 * ib_dealloc_xrcd - Deallocates an XRC domain. 3883 * @xrcd: The XRC domain to deallocate. 3884 */ 3885 int ib_dealloc_xrcd(struct ib_xrcd *xrcd); 3886 3887 static inline int ib_check_mr_access(int flags) 3888 { 3889 /* 3890 * Local write permission is required if remote write or 3891 * remote atomic permission is also requested. 3892 */ 3893 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) && 3894 !(flags & IB_ACCESS_LOCAL_WRITE)) 3895 return -EINVAL; 3896 3897 return 0; 3898 } 3899 3900 static inline bool ib_access_writable(int access_flags) 3901 { 3902 /* 3903 * We have writable memory backing the MR if any of the following 3904 * access flags are set. "Local write" and "remote write" obviously 3905 * require write access. "Remote atomic" can do things like fetch and 3906 * add, which will modify memory, and "MW bind" can change permissions 3907 * by binding a window. 3908 */ 3909 return access_flags & 3910 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE | 3911 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND); 3912 } 3913 3914 /** 3915 * ib_check_mr_status: lightweight check of MR status. 3916 * This routine may provide status checks on a selected 3917 * ib_mr. first use is for signature status check. 3918 * 3919 * @mr: A memory region. 3920 * @check_mask: Bitmask of which checks to perform from 3921 * ib_mr_status_check enumeration. 3922 * @mr_status: The container of relevant status checks. 3923 * failed checks will be indicated in the status bitmask 3924 * and the relevant info shall be in the error item. 3925 */ 3926 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask, 3927 struct ib_mr_status *mr_status); 3928 3929 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port, 3930 u16 pkey, const union ib_gid *gid, 3931 const struct sockaddr *addr); 3932 struct ib_wq *ib_create_wq(struct ib_pd *pd, 3933 struct ib_wq_init_attr *init_attr); 3934 int ib_destroy_wq(struct ib_wq *wq); 3935 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr, 3936 u32 wq_attr_mask); 3937 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device, 3938 struct ib_rwq_ind_table_init_attr* 3939 wq_ind_table_init_attr); 3940 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table); 3941 3942 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents, 3943 unsigned int *sg_offset, unsigned int page_size); 3944 3945 static inline int 3946 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents, 3947 unsigned int *sg_offset, unsigned int page_size) 3948 { 3949 int n; 3950 3951 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size); 3952 mr->iova = 0; 3953 3954 return n; 3955 } 3956 3957 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents, 3958 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64)); 3959 3960 void ib_drain_rq(struct ib_qp *qp); 3961 void ib_drain_sq(struct ib_qp *qp); 3962 void ib_drain_qp(struct ib_qp *qp); 3963 3964 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width); 3965 3966 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr) 3967 { 3968 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE) 3969 return attr->roce.dmac; 3970 return NULL; 3971 } 3972 3973 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid) 3974 { 3975 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 3976 attr->ib.dlid = (u16)dlid; 3977 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 3978 attr->opa.dlid = dlid; 3979 } 3980 3981 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr) 3982 { 3983 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 3984 return attr->ib.dlid; 3985 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 3986 return attr->opa.dlid; 3987 return 0; 3988 } 3989 3990 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl) 3991 { 3992 attr->sl = sl; 3993 } 3994 3995 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr) 3996 { 3997 return attr->sl; 3998 } 3999 4000 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr, 4001 u8 src_path_bits) 4002 { 4003 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 4004 attr->ib.src_path_bits = src_path_bits; 4005 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 4006 attr->opa.src_path_bits = src_path_bits; 4007 } 4008 4009 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr) 4010 { 4011 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 4012 return attr->ib.src_path_bits; 4013 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 4014 return attr->opa.src_path_bits; 4015 return 0; 4016 } 4017 4018 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr, 4019 bool make_grd) 4020 { 4021 if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 4022 attr->opa.make_grd = make_grd; 4023 } 4024 4025 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr) 4026 { 4027 if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 4028 return attr->opa.make_grd; 4029 return false; 4030 } 4031 4032 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num) 4033 { 4034 attr->port_num = port_num; 4035 } 4036 4037 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr) 4038 { 4039 return attr->port_num; 4040 } 4041 4042 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr, 4043 u8 static_rate) 4044 { 4045 attr->static_rate = static_rate; 4046 } 4047 4048 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr) 4049 { 4050 return attr->static_rate; 4051 } 4052 4053 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr, 4054 enum ib_ah_flags flag) 4055 { 4056 attr->ah_flags = flag; 4057 } 4058 4059 static inline enum ib_ah_flags 4060 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr) 4061 { 4062 return attr->ah_flags; 4063 } 4064 4065 static inline const struct ib_global_route 4066 *rdma_ah_read_grh(const struct rdma_ah_attr *attr) 4067 { 4068 return &attr->grh; 4069 } 4070 4071 /*To retrieve and modify the grh */ 4072 static inline struct ib_global_route 4073 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr) 4074 { 4075 return &attr->grh; 4076 } 4077 4078 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid) 4079 { 4080 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 4081 4082 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid)); 4083 } 4084 4085 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr, 4086 __be64 prefix) 4087 { 4088 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 4089 4090 grh->dgid.global.subnet_prefix = prefix; 4091 } 4092 4093 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr, 4094 __be64 if_id) 4095 { 4096 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 4097 4098 grh->dgid.global.interface_id = if_id; 4099 } 4100 4101 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr, 4102 union ib_gid *dgid, u32 flow_label, 4103 u8 sgid_index, u8 hop_limit, 4104 u8 traffic_class) 4105 { 4106 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 4107 4108 attr->ah_flags = IB_AH_GRH; 4109 if (dgid) 4110 grh->dgid = *dgid; 4111 grh->flow_label = flow_label; 4112 grh->sgid_index = sgid_index; 4113 grh->hop_limit = hop_limit; 4114 grh->traffic_class = traffic_class; 4115 grh->sgid_attr = NULL; 4116 } 4117 4118 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr); 4119 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid, 4120 u32 flow_label, u8 hop_limit, u8 traffic_class, 4121 const struct ib_gid_attr *sgid_attr); 4122 void rdma_copy_ah_attr(struct rdma_ah_attr *dest, 4123 const struct rdma_ah_attr *src); 4124 void rdma_replace_ah_attr(struct rdma_ah_attr *old, 4125 const struct rdma_ah_attr *new); 4126 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src); 4127 4128 /** 4129 * rdma_ah_find_type - Return address handle type. 4130 * 4131 * @dev: Device to be checked 4132 * @port_num: Port number 4133 */ 4134 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev, 4135 u8 port_num) 4136 { 4137 if (rdma_protocol_roce(dev, port_num)) 4138 return RDMA_AH_ATTR_TYPE_ROCE; 4139 if (rdma_protocol_ib(dev, port_num)) { 4140 if (rdma_cap_opa_ah(dev, port_num)) 4141 return RDMA_AH_ATTR_TYPE_OPA; 4142 return RDMA_AH_ATTR_TYPE_IB; 4143 } 4144 4145 return RDMA_AH_ATTR_TYPE_UNDEFINED; 4146 } 4147 4148 /** 4149 * ib_lid_cpu16 - Return lid in 16bit CPU encoding. 4150 * In the current implementation the only way to get 4151 * get the 32bit lid is from other sources for OPA. 4152 * For IB, lids will always be 16bits so cast the 4153 * value accordingly. 4154 * 4155 * @lid: A 32bit LID 4156 */ 4157 static inline u16 ib_lid_cpu16(u32 lid) 4158 { 4159 WARN_ON_ONCE(lid & 0xFFFF0000); 4160 return (u16)lid; 4161 } 4162 4163 /** 4164 * ib_lid_be16 - Return lid in 16bit BE encoding. 4165 * 4166 * @lid: A 32bit LID 4167 */ 4168 static inline __be16 ib_lid_be16(u32 lid) 4169 { 4170 WARN_ON_ONCE(lid & 0xFFFF0000); 4171 return cpu_to_be16((u16)lid); 4172 } 4173 4174 /** 4175 * ib_get_vector_affinity - Get the affinity mappings of a given completion 4176 * vector 4177 * @device: the rdma device 4178 * @comp_vector: index of completion vector 4179 * 4180 * Returns NULL on failure, otherwise a corresponding cpu map of the 4181 * completion vector (returns all-cpus map if the device driver doesn't 4182 * implement get_vector_affinity). 4183 */ 4184 static inline const struct cpumask * 4185 ib_get_vector_affinity(struct ib_device *device, int comp_vector) 4186 { 4187 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors || 4188 !device->ops.get_vector_affinity) 4189 return NULL; 4190 4191 return device->ops.get_vector_affinity(device, comp_vector); 4192 4193 } 4194 4195 /** 4196 * rdma_roce_rescan_device - Rescan all of the network devices in the system 4197 * and add their gids, as needed, to the relevant RoCE devices. 4198 * 4199 * @device: the rdma device 4200 */ 4201 void rdma_roce_rescan_device(struct ib_device *ibdev); 4202 4203 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile); 4204 4205 4206 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs); 4207 4208 struct net_device *rdma_alloc_netdev(struct ib_device *device, u8 port_num, 4209 enum rdma_netdev_t type, const char *name, 4210 unsigned char name_assign_type, 4211 void (*setup)(struct net_device *)); 4212 4213 int rdma_init_netdev(struct ib_device *device, u8 port_num, 4214 enum rdma_netdev_t type, const char *name, 4215 unsigned char name_assign_type, 4216 void (*setup)(struct net_device *), 4217 struct net_device *netdev); 4218 4219 /** 4220 * rdma_set_device_sysfs_group - Set device attributes group to have 4221 * driver specific sysfs entries at 4222 * for infiniband class. 4223 * 4224 * @device: device pointer for which attributes to be created 4225 * @group: Pointer to group which should be added when device 4226 * is registered with sysfs. 4227 * rdma_set_device_sysfs_group() allows existing drivers to expose one 4228 * group per device to have sysfs attributes. 4229 * 4230 * NOTE: New drivers should not make use of this API; instead new device 4231 * parameter should be exposed via netlink command. This API and mechanism 4232 * exist only for existing drivers. 4233 */ 4234 static inline void 4235 rdma_set_device_sysfs_group(struct ib_device *dev, 4236 const struct attribute_group *group) 4237 { 4238 dev->groups[1] = group; 4239 } 4240 4241 #endif /* IB_VERBS_H */ 4242