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 <linux/irqflags.h> 63 #include <linux/preempt.h> 64 #include <uapi/rdma/ib_user_verbs.h> 65 #include <rdma/restrack.h> 66 #include <uapi/rdma/rdma_user_ioctl.h> 67 #include <uapi/rdma/ib_user_ioctl_verbs.h> 68 69 #define IB_FW_VERSION_NAME_MAX ETHTOOL_FWVERS_LEN 70 71 struct ib_umem_odp; 72 73 extern struct workqueue_struct *ib_wq; 74 extern struct workqueue_struct *ib_comp_wq; 75 extern struct workqueue_struct *ib_comp_unbound_wq; 76 77 __printf(3, 4) __cold 78 void ibdev_printk(const char *level, const struct ib_device *ibdev, 79 const char *format, ...); 80 __printf(2, 3) __cold 81 void ibdev_emerg(const struct ib_device *ibdev, const char *format, ...); 82 __printf(2, 3) __cold 83 void ibdev_alert(const struct ib_device *ibdev, const char *format, ...); 84 __printf(2, 3) __cold 85 void ibdev_crit(const struct ib_device *ibdev, const char *format, ...); 86 __printf(2, 3) __cold 87 void ibdev_err(const struct ib_device *ibdev, const char *format, ...); 88 __printf(2, 3) __cold 89 void ibdev_warn(const struct ib_device *ibdev, const char *format, ...); 90 __printf(2, 3) __cold 91 void ibdev_notice(const struct ib_device *ibdev, const char *format, ...); 92 __printf(2, 3) __cold 93 void ibdev_info(const struct ib_device *ibdev, const char *format, ...); 94 95 #if defined(CONFIG_DYNAMIC_DEBUG) 96 #define ibdev_dbg(__dev, format, args...) \ 97 dynamic_ibdev_dbg(__dev, format, ##args) 98 #elif defined(DEBUG) 99 #define ibdev_dbg(__dev, format, args...) \ 100 ibdev_printk(KERN_DEBUG, __dev, format, ##args) 101 #else 102 __printf(2, 3) __cold 103 static inline 104 void ibdev_dbg(const struct ib_device *ibdev, const char *format, ...) {} 105 #endif 106 107 union ib_gid { 108 u8 raw[16]; 109 struct { 110 __be64 subnet_prefix; 111 __be64 interface_id; 112 } global; 113 }; 114 115 extern union ib_gid zgid; 116 117 enum ib_gid_type { 118 /* If link layer is Ethernet, this is RoCE V1 */ 119 IB_GID_TYPE_IB = 0, 120 IB_GID_TYPE_ROCE = 0, 121 IB_GID_TYPE_ROCE_UDP_ENCAP = 1, 122 IB_GID_TYPE_SIZE 123 }; 124 125 #define ROCE_V2_UDP_DPORT 4791 126 struct ib_gid_attr { 127 struct net_device __rcu *ndev; 128 struct ib_device *device; 129 union ib_gid gid; 130 enum ib_gid_type gid_type; 131 u16 index; 132 u8 port_num; 133 }; 134 135 enum rdma_node_type { 136 /* IB values map to NodeInfo:NodeType. */ 137 RDMA_NODE_IB_CA = 1, 138 RDMA_NODE_IB_SWITCH, 139 RDMA_NODE_IB_ROUTER, 140 RDMA_NODE_RNIC, 141 RDMA_NODE_USNIC, 142 RDMA_NODE_USNIC_UDP, 143 RDMA_NODE_UNSPECIFIED, 144 }; 145 146 enum { 147 /* set the local administered indication */ 148 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2, 149 }; 150 151 enum rdma_transport_type { 152 RDMA_TRANSPORT_IB, 153 RDMA_TRANSPORT_IWARP, 154 RDMA_TRANSPORT_USNIC, 155 RDMA_TRANSPORT_USNIC_UDP, 156 RDMA_TRANSPORT_UNSPECIFIED, 157 }; 158 159 enum rdma_protocol_type { 160 RDMA_PROTOCOL_IB, 161 RDMA_PROTOCOL_IBOE, 162 RDMA_PROTOCOL_IWARP, 163 RDMA_PROTOCOL_USNIC_UDP 164 }; 165 166 __attribute_const__ enum rdma_transport_type 167 rdma_node_get_transport(enum rdma_node_type node_type); 168 169 enum rdma_network_type { 170 RDMA_NETWORK_IB, 171 RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB, 172 RDMA_NETWORK_IPV4, 173 RDMA_NETWORK_IPV6 174 }; 175 176 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type) 177 { 178 if (network_type == RDMA_NETWORK_IPV4 || 179 network_type == RDMA_NETWORK_IPV6) 180 return IB_GID_TYPE_ROCE_UDP_ENCAP; 181 182 /* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */ 183 return IB_GID_TYPE_IB; 184 } 185 186 static inline enum rdma_network_type 187 rdma_gid_attr_network_type(const struct ib_gid_attr *attr) 188 { 189 if (attr->gid_type == IB_GID_TYPE_IB) 190 return RDMA_NETWORK_IB; 191 192 if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid)) 193 return RDMA_NETWORK_IPV4; 194 else 195 return RDMA_NETWORK_IPV6; 196 } 197 198 enum rdma_link_layer { 199 IB_LINK_LAYER_UNSPECIFIED, 200 IB_LINK_LAYER_INFINIBAND, 201 IB_LINK_LAYER_ETHERNET, 202 }; 203 204 enum ib_device_cap_flags { 205 IB_DEVICE_RESIZE_MAX_WR = (1 << 0), 206 IB_DEVICE_BAD_PKEY_CNTR = (1 << 1), 207 IB_DEVICE_BAD_QKEY_CNTR = (1 << 2), 208 IB_DEVICE_RAW_MULTI = (1 << 3), 209 IB_DEVICE_AUTO_PATH_MIG = (1 << 4), 210 IB_DEVICE_CHANGE_PHY_PORT = (1 << 5), 211 IB_DEVICE_UD_AV_PORT_ENFORCE = (1 << 6), 212 IB_DEVICE_CURR_QP_STATE_MOD = (1 << 7), 213 IB_DEVICE_SHUTDOWN_PORT = (1 << 8), 214 /* Not in use, former INIT_TYPE = (1 << 9),*/ 215 IB_DEVICE_PORT_ACTIVE_EVENT = (1 << 10), 216 IB_DEVICE_SYS_IMAGE_GUID = (1 << 11), 217 IB_DEVICE_RC_RNR_NAK_GEN = (1 << 12), 218 IB_DEVICE_SRQ_RESIZE = (1 << 13), 219 IB_DEVICE_N_NOTIFY_CQ = (1 << 14), 220 221 /* 222 * This device supports a per-device lkey or stag that can be 223 * used without performing a memory registration for the local 224 * memory. Note that ULPs should never check this flag, but 225 * instead of use the local_dma_lkey flag in the ib_pd structure, 226 * which will always contain a usable lkey. 227 */ 228 IB_DEVICE_LOCAL_DMA_LKEY = (1 << 15), 229 /* Reserved, old SEND_W_INV = (1 << 16),*/ 230 IB_DEVICE_MEM_WINDOW = (1 << 17), 231 /* 232 * Devices should set IB_DEVICE_UD_IP_SUM if they support 233 * insertion of UDP and TCP checksum on outgoing UD IPoIB 234 * messages and can verify the validity of checksum for 235 * incoming messages. Setting this flag implies that the 236 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode. 237 */ 238 IB_DEVICE_UD_IP_CSUM = (1 << 18), 239 IB_DEVICE_UD_TSO = (1 << 19), 240 IB_DEVICE_XRC = (1 << 20), 241 242 /* 243 * This device supports the IB "base memory management extension", 244 * which includes support for fast registrations (IB_WR_REG_MR, 245 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should 246 * also be set by any iWarp device which must support FRs to comply 247 * to the iWarp verbs spec. iWarp devices also support the 248 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the 249 * stag. 250 */ 251 IB_DEVICE_MEM_MGT_EXTENSIONS = (1 << 21), 252 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1 << 22), 253 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1 << 23), 254 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1 << 24), 255 IB_DEVICE_RC_IP_CSUM = (1 << 25), 256 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */ 257 IB_DEVICE_RAW_IP_CSUM = (1 << 26), 258 /* 259 * Devices should set IB_DEVICE_CROSS_CHANNEL if they 260 * support execution of WQEs that involve synchronization 261 * of I/O operations with single completion queue managed 262 * by hardware. 263 */ 264 IB_DEVICE_CROSS_CHANNEL = (1 << 27), 265 IB_DEVICE_MANAGED_FLOW_STEERING = (1 << 29), 266 IB_DEVICE_SIGNATURE_HANDOVER = (1 << 30), 267 IB_DEVICE_ON_DEMAND_PAGING = (1ULL << 31), 268 IB_DEVICE_SG_GAPS_REG = (1ULL << 32), 269 IB_DEVICE_VIRTUAL_FUNCTION = (1ULL << 33), 270 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */ 271 IB_DEVICE_RAW_SCATTER_FCS = (1ULL << 34), 272 IB_DEVICE_RDMA_NETDEV_OPA_VNIC = (1ULL << 35), 273 /* The device supports padding incoming writes to cacheline. */ 274 IB_DEVICE_PCI_WRITE_END_PADDING = (1ULL << 36), 275 IB_DEVICE_ALLOW_USER_UNREG = (1ULL << 37), 276 }; 277 278 enum ib_signature_prot_cap { 279 IB_PROT_T10DIF_TYPE_1 = 1, 280 IB_PROT_T10DIF_TYPE_2 = 1 << 1, 281 IB_PROT_T10DIF_TYPE_3 = 1 << 2, 282 }; 283 284 enum ib_signature_guard_cap { 285 IB_GUARD_T10DIF_CRC = 1, 286 IB_GUARD_T10DIF_CSUM = 1 << 1, 287 }; 288 289 enum ib_atomic_cap { 290 IB_ATOMIC_NONE, 291 IB_ATOMIC_HCA, 292 IB_ATOMIC_GLOB 293 }; 294 295 enum ib_odp_general_cap_bits { 296 IB_ODP_SUPPORT = 1 << 0, 297 IB_ODP_SUPPORT_IMPLICIT = 1 << 1, 298 }; 299 300 enum ib_odp_transport_cap_bits { 301 IB_ODP_SUPPORT_SEND = 1 << 0, 302 IB_ODP_SUPPORT_RECV = 1 << 1, 303 IB_ODP_SUPPORT_WRITE = 1 << 2, 304 IB_ODP_SUPPORT_READ = 1 << 3, 305 IB_ODP_SUPPORT_ATOMIC = 1 << 4, 306 IB_ODP_SUPPORT_SRQ_RECV = 1 << 5, 307 }; 308 309 struct ib_odp_caps { 310 uint64_t general_caps; 311 struct { 312 uint32_t rc_odp_caps; 313 uint32_t uc_odp_caps; 314 uint32_t ud_odp_caps; 315 uint32_t xrc_odp_caps; 316 } per_transport_caps; 317 }; 318 319 struct ib_rss_caps { 320 /* Corresponding bit will be set if qp type from 321 * 'enum ib_qp_type' is supported, e.g. 322 * supported_qpts |= 1 << IB_QPT_UD 323 */ 324 u32 supported_qpts; 325 u32 max_rwq_indirection_tables; 326 u32 max_rwq_indirection_table_size; 327 }; 328 329 enum ib_tm_cap_flags { 330 /* Support tag matching on RC transport */ 331 IB_TM_CAP_RC = 1 << 0, 332 }; 333 334 struct ib_tm_caps { 335 /* Max size of RNDV header */ 336 u32 max_rndv_hdr_size; 337 /* Max number of entries in tag matching list */ 338 u32 max_num_tags; 339 /* From enum ib_tm_cap_flags */ 340 u32 flags; 341 /* Max number of outstanding list operations */ 342 u32 max_ops; 343 /* Max number of SGE in tag matching entry */ 344 u32 max_sge; 345 }; 346 347 struct ib_cq_init_attr { 348 unsigned int cqe; 349 int comp_vector; 350 u32 flags; 351 }; 352 353 enum ib_cq_attr_mask { 354 IB_CQ_MODERATE = 1 << 0, 355 }; 356 357 struct ib_cq_caps { 358 u16 max_cq_moderation_count; 359 u16 max_cq_moderation_period; 360 }; 361 362 struct ib_dm_mr_attr { 363 u64 length; 364 u64 offset; 365 u32 access_flags; 366 }; 367 368 struct ib_dm_alloc_attr { 369 u64 length; 370 u32 alignment; 371 u32 flags; 372 }; 373 374 struct ib_device_attr { 375 u64 fw_ver; 376 __be64 sys_image_guid; 377 u64 max_mr_size; 378 u64 page_size_cap; 379 u32 vendor_id; 380 u32 vendor_part_id; 381 u32 hw_ver; 382 int max_qp; 383 int max_qp_wr; 384 u64 device_cap_flags; 385 int max_send_sge; 386 int max_recv_sge; 387 int max_sge_rd; 388 int max_cq; 389 int max_cqe; 390 int max_mr; 391 int max_pd; 392 int max_qp_rd_atom; 393 int max_ee_rd_atom; 394 int max_res_rd_atom; 395 int max_qp_init_rd_atom; 396 int max_ee_init_rd_atom; 397 enum ib_atomic_cap atomic_cap; 398 enum ib_atomic_cap masked_atomic_cap; 399 int max_ee; 400 int max_rdd; 401 int max_mw; 402 int max_raw_ipv6_qp; 403 int max_raw_ethy_qp; 404 int max_mcast_grp; 405 int max_mcast_qp_attach; 406 int max_total_mcast_qp_attach; 407 int max_ah; 408 int max_fmr; 409 int max_map_per_fmr; 410 int max_srq; 411 int max_srq_wr; 412 int max_srq_sge; 413 unsigned int max_fast_reg_page_list_len; 414 u16 max_pkeys; 415 u8 local_ca_ack_delay; 416 int sig_prot_cap; 417 int sig_guard_cap; 418 struct ib_odp_caps odp_caps; 419 uint64_t timestamp_mask; 420 uint64_t hca_core_clock; /* in KHZ */ 421 struct ib_rss_caps rss_caps; 422 u32 max_wq_type_rq; 423 u32 raw_packet_caps; /* Use ib_raw_packet_caps enum */ 424 struct ib_tm_caps tm_caps; 425 struct ib_cq_caps cq_caps; 426 u64 max_dm_size; 427 }; 428 429 enum ib_mtu { 430 IB_MTU_256 = 1, 431 IB_MTU_512 = 2, 432 IB_MTU_1024 = 3, 433 IB_MTU_2048 = 4, 434 IB_MTU_4096 = 5 435 }; 436 437 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu) 438 { 439 switch (mtu) { 440 case IB_MTU_256: return 256; 441 case IB_MTU_512: return 512; 442 case IB_MTU_1024: return 1024; 443 case IB_MTU_2048: return 2048; 444 case IB_MTU_4096: return 4096; 445 default: return -1; 446 } 447 } 448 449 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu) 450 { 451 if (mtu >= 4096) 452 return IB_MTU_4096; 453 else if (mtu >= 2048) 454 return IB_MTU_2048; 455 else if (mtu >= 1024) 456 return IB_MTU_1024; 457 else if (mtu >= 512) 458 return IB_MTU_512; 459 else 460 return IB_MTU_256; 461 } 462 463 enum ib_port_state { 464 IB_PORT_NOP = 0, 465 IB_PORT_DOWN = 1, 466 IB_PORT_INIT = 2, 467 IB_PORT_ARMED = 3, 468 IB_PORT_ACTIVE = 4, 469 IB_PORT_ACTIVE_DEFER = 5 470 }; 471 472 enum ib_port_width { 473 IB_WIDTH_1X = 1, 474 IB_WIDTH_2X = 16, 475 IB_WIDTH_4X = 2, 476 IB_WIDTH_8X = 4, 477 IB_WIDTH_12X = 8 478 }; 479 480 static inline int ib_width_enum_to_int(enum ib_port_width width) 481 { 482 switch (width) { 483 case IB_WIDTH_1X: return 1; 484 case IB_WIDTH_2X: return 2; 485 case IB_WIDTH_4X: return 4; 486 case IB_WIDTH_8X: return 8; 487 case IB_WIDTH_12X: return 12; 488 default: return -1; 489 } 490 } 491 492 enum ib_port_speed { 493 IB_SPEED_SDR = 1, 494 IB_SPEED_DDR = 2, 495 IB_SPEED_QDR = 4, 496 IB_SPEED_FDR10 = 8, 497 IB_SPEED_FDR = 16, 498 IB_SPEED_EDR = 32, 499 IB_SPEED_HDR = 64 500 }; 501 502 /** 503 * struct rdma_hw_stats 504 * @lock - Mutex to protect parallel write access to lifespan and values 505 * of counters, which are 64bits and not guaranteeed to be written 506 * atomicaly on 32bits systems. 507 * @timestamp - Used by the core code to track when the last update was 508 * @lifespan - Used by the core code to determine how old the counters 509 * should be before being updated again. Stored in jiffies, defaults 510 * to 10 milliseconds, drivers can override the default be specifying 511 * their own value during their allocation routine. 512 * @name - Array of pointers to static names used for the counters in 513 * directory. 514 * @num_counters - How many hardware counters there are. If name is 515 * shorter than this number, a kernel oops will result. Driver authors 516 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters) 517 * in their code to prevent this. 518 * @value - Array of u64 counters that are accessed by the sysfs code and 519 * filled in by the drivers get_stats routine 520 */ 521 struct rdma_hw_stats { 522 struct mutex lock; /* Protect lifespan and values[] */ 523 unsigned long timestamp; 524 unsigned long lifespan; 525 const char * const *names; 526 int num_counters; 527 u64 value[]; 528 }; 529 530 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10 531 /** 532 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct 533 * for drivers. 534 * @names - Array of static const char * 535 * @num_counters - How many elements in array 536 * @lifespan - How many milliseconds between updates 537 */ 538 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct( 539 const char * const *names, int num_counters, 540 unsigned long lifespan) 541 { 542 struct rdma_hw_stats *stats; 543 544 stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64), 545 GFP_KERNEL); 546 if (!stats) 547 return NULL; 548 stats->names = names; 549 stats->num_counters = num_counters; 550 stats->lifespan = msecs_to_jiffies(lifespan); 551 552 return stats; 553 } 554 555 556 /* Define bits for the various functionality this port needs to be supported by 557 * the core. 558 */ 559 /* Management 0x00000FFF */ 560 #define RDMA_CORE_CAP_IB_MAD 0x00000001 561 #define RDMA_CORE_CAP_IB_SMI 0x00000002 562 #define RDMA_CORE_CAP_IB_CM 0x00000004 563 #define RDMA_CORE_CAP_IW_CM 0x00000008 564 #define RDMA_CORE_CAP_IB_SA 0x00000010 565 #define RDMA_CORE_CAP_OPA_MAD 0x00000020 566 567 /* Address format 0x000FF000 */ 568 #define RDMA_CORE_CAP_AF_IB 0x00001000 569 #define RDMA_CORE_CAP_ETH_AH 0x00002000 570 #define RDMA_CORE_CAP_OPA_AH 0x00004000 571 #define RDMA_CORE_CAP_IB_GRH_REQUIRED 0x00008000 572 573 /* Protocol 0xFFF00000 */ 574 #define RDMA_CORE_CAP_PROT_IB 0x00100000 575 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000 576 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000 577 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000 578 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000 579 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000 580 581 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \ 582 | RDMA_CORE_CAP_PROT_ROCE \ 583 | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP) 584 585 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \ 586 | RDMA_CORE_CAP_IB_MAD \ 587 | RDMA_CORE_CAP_IB_SMI \ 588 | RDMA_CORE_CAP_IB_CM \ 589 | RDMA_CORE_CAP_IB_SA \ 590 | RDMA_CORE_CAP_AF_IB) 591 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \ 592 | RDMA_CORE_CAP_IB_MAD \ 593 | RDMA_CORE_CAP_IB_CM \ 594 | RDMA_CORE_CAP_AF_IB \ 595 | RDMA_CORE_CAP_ETH_AH) 596 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \ 597 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \ 598 | RDMA_CORE_CAP_IB_MAD \ 599 | RDMA_CORE_CAP_IB_CM \ 600 | RDMA_CORE_CAP_AF_IB \ 601 | RDMA_CORE_CAP_ETH_AH) 602 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \ 603 | RDMA_CORE_CAP_IW_CM) 604 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \ 605 | RDMA_CORE_CAP_OPA_MAD) 606 607 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET) 608 609 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC) 610 611 struct ib_port_attr { 612 u64 subnet_prefix; 613 enum ib_port_state state; 614 enum ib_mtu max_mtu; 615 enum ib_mtu active_mtu; 616 int gid_tbl_len; 617 unsigned int ip_gids:1; 618 /* This is the value from PortInfo CapabilityMask, defined by IBA */ 619 u32 port_cap_flags; 620 u32 max_msg_sz; 621 u32 bad_pkey_cntr; 622 u32 qkey_viol_cntr; 623 u16 pkey_tbl_len; 624 u32 sm_lid; 625 u32 lid; 626 u8 lmc; 627 u8 max_vl_num; 628 u8 sm_sl; 629 u8 subnet_timeout; 630 u8 init_type_reply; 631 u8 active_width; 632 u8 active_speed; 633 u8 phys_state; 634 u16 port_cap_flags2; 635 }; 636 637 enum ib_device_modify_flags { 638 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0, 639 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1 640 }; 641 642 #define IB_DEVICE_NODE_DESC_MAX 64 643 644 struct ib_device_modify { 645 u64 sys_image_guid; 646 char node_desc[IB_DEVICE_NODE_DESC_MAX]; 647 }; 648 649 enum ib_port_modify_flags { 650 IB_PORT_SHUTDOWN = 1, 651 IB_PORT_INIT_TYPE = (1<<2), 652 IB_PORT_RESET_QKEY_CNTR = (1<<3), 653 IB_PORT_OPA_MASK_CHG = (1<<4) 654 }; 655 656 struct ib_port_modify { 657 u32 set_port_cap_mask; 658 u32 clr_port_cap_mask; 659 u8 init_type; 660 }; 661 662 enum ib_event_type { 663 IB_EVENT_CQ_ERR, 664 IB_EVENT_QP_FATAL, 665 IB_EVENT_QP_REQ_ERR, 666 IB_EVENT_QP_ACCESS_ERR, 667 IB_EVENT_COMM_EST, 668 IB_EVENT_SQ_DRAINED, 669 IB_EVENT_PATH_MIG, 670 IB_EVENT_PATH_MIG_ERR, 671 IB_EVENT_DEVICE_FATAL, 672 IB_EVENT_PORT_ACTIVE, 673 IB_EVENT_PORT_ERR, 674 IB_EVENT_LID_CHANGE, 675 IB_EVENT_PKEY_CHANGE, 676 IB_EVENT_SM_CHANGE, 677 IB_EVENT_SRQ_ERR, 678 IB_EVENT_SRQ_LIMIT_REACHED, 679 IB_EVENT_QP_LAST_WQE_REACHED, 680 IB_EVENT_CLIENT_REREGISTER, 681 IB_EVENT_GID_CHANGE, 682 IB_EVENT_WQ_FATAL, 683 }; 684 685 const char *__attribute_const__ ib_event_msg(enum ib_event_type event); 686 687 struct ib_event { 688 struct ib_device *device; 689 union { 690 struct ib_cq *cq; 691 struct ib_qp *qp; 692 struct ib_srq *srq; 693 struct ib_wq *wq; 694 u8 port_num; 695 } element; 696 enum ib_event_type event; 697 }; 698 699 struct ib_event_handler { 700 struct ib_device *device; 701 void (*handler)(struct ib_event_handler *, struct ib_event *); 702 struct list_head list; 703 }; 704 705 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \ 706 do { \ 707 (_ptr)->device = _device; \ 708 (_ptr)->handler = _handler; \ 709 INIT_LIST_HEAD(&(_ptr)->list); \ 710 } while (0) 711 712 struct ib_global_route { 713 const struct ib_gid_attr *sgid_attr; 714 union ib_gid dgid; 715 u32 flow_label; 716 u8 sgid_index; 717 u8 hop_limit; 718 u8 traffic_class; 719 }; 720 721 struct ib_grh { 722 __be32 version_tclass_flow; 723 __be16 paylen; 724 u8 next_hdr; 725 u8 hop_limit; 726 union ib_gid sgid; 727 union ib_gid dgid; 728 }; 729 730 union rdma_network_hdr { 731 struct ib_grh ibgrh; 732 struct { 733 /* The IB spec states that if it's IPv4, the header 734 * is located in the last 20 bytes of the header. 735 */ 736 u8 reserved[20]; 737 struct iphdr roce4grh; 738 }; 739 }; 740 741 #define IB_QPN_MASK 0xFFFFFF 742 743 enum { 744 IB_MULTICAST_QPN = 0xffffff 745 }; 746 747 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF) 748 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000) 749 750 enum ib_ah_flags { 751 IB_AH_GRH = 1 752 }; 753 754 enum ib_rate { 755 IB_RATE_PORT_CURRENT = 0, 756 IB_RATE_2_5_GBPS = 2, 757 IB_RATE_5_GBPS = 5, 758 IB_RATE_10_GBPS = 3, 759 IB_RATE_20_GBPS = 6, 760 IB_RATE_30_GBPS = 4, 761 IB_RATE_40_GBPS = 7, 762 IB_RATE_60_GBPS = 8, 763 IB_RATE_80_GBPS = 9, 764 IB_RATE_120_GBPS = 10, 765 IB_RATE_14_GBPS = 11, 766 IB_RATE_56_GBPS = 12, 767 IB_RATE_112_GBPS = 13, 768 IB_RATE_168_GBPS = 14, 769 IB_RATE_25_GBPS = 15, 770 IB_RATE_100_GBPS = 16, 771 IB_RATE_200_GBPS = 17, 772 IB_RATE_300_GBPS = 18, 773 IB_RATE_28_GBPS = 19, 774 IB_RATE_50_GBPS = 20, 775 IB_RATE_400_GBPS = 21, 776 IB_RATE_600_GBPS = 22, 777 }; 778 779 /** 780 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the 781 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be 782 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec. 783 * @rate: rate to convert. 784 */ 785 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate); 786 787 /** 788 * ib_rate_to_mbps - Convert the IB rate enum to Mbps. 789 * For example, IB_RATE_2_5_GBPS will be converted to 2500. 790 * @rate: rate to convert. 791 */ 792 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate); 793 794 795 /** 796 * enum ib_mr_type - memory region type 797 * @IB_MR_TYPE_MEM_REG: memory region that is used for 798 * normal registration 799 * @IB_MR_TYPE_SIGNATURE: memory region that is used for 800 * signature operations (data-integrity 801 * capable regions) 802 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to 803 * register any arbitrary sg lists (without 804 * the normal mr constraints - see 805 * ib_map_mr_sg) 806 */ 807 enum ib_mr_type { 808 IB_MR_TYPE_MEM_REG, 809 IB_MR_TYPE_SIGNATURE, 810 IB_MR_TYPE_SG_GAPS, 811 }; 812 813 /** 814 * Signature types 815 * IB_SIG_TYPE_NONE: Unprotected. 816 * IB_SIG_TYPE_T10_DIF: Type T10-DIF 817 */ 818 enum ib_signature_type { 819 IB_SIG_TYPE_NONE, 820 IB_SIG_TYPE_T10_DIF, 821 }; 822 823 /** 824 * Signature T10-DIF block-guard types 825 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules. 826 * IB_T10DIF_CSUM: Corresponds to IP checksum rules. 827 */ 828 enum ib_t10_dif_bg_type { 829 IB_T10DIF_CRC, 830 IB_T10DIF_CSUM 831 }; 832 833 /** 834 * struct ib_t10_dif_domain - Parameters specific for T10-DIF 835 * domain. 836 * @bg_type: T10-DIF block guard type (CRC|CSUM) 837 * @pi_interval: protection information interval. 838 * @bg: seed of guard computation. 839 * @app_tag: application tag of guard block 840 * @ref_tag: initial guard block reference tag. 841 * @ref_remap: Indicate wethear the reftag increments each block 842 * @app_escape: Indicate to skip block check if apptag=0xffff 843 * @ref_escape: Indicate to skip block check if reftag=0xffffffff 844 * @apptag_check_mask: check bitmask of application tag. 845 */ 846 struct ib_t10_dif_domain { 847 enum ib_t10_dif_bg_type bg_type; 848 u16 pi_interval; 849 u16 bg; 850 u16 app_tag; 851 u32 ref_tag; 852 bool ref_remap; 853 bool app_escape; 854 bool ref_escape; 855 u16 apptag_check_mask; 856 }; 857 858 /** 859 * struct ib_sig_domain - Parameters for signature domain 860 * @sig_type: specific signauture type 861 * @sig: union of all signature domain attributes that may 862 * be used to set domain layout. 863 */ 864 struct ib_sig_domain { 865 enum ib_signature_type sig_type; 866 union { 867 struct ib_t10_dif_domain dif; 868 } sig; 869 }; 870 871 /** 872 * struct ib_sig_attrs - Parameters for signature handover operation 873 * @check_mask: bitmask for signature byte check (8 bytes) 874 * @mem: memory domain layout desciptor. 875 * @wire: wire domain layout desciptor. 876 */ 877 struct ib_sig_attrs { 878 u8 check_mask; 879 struct ib_sig_domain mem; 880 struct ib_sig_domain wire; 881 }; 882 883 enum ib_sig_err_type { 884 IB_SIG_BAD_GUARD, 885 IB_SIG_BAD_REFTAG, 886 IB_SIG_BAD_APPTAG, 887 }; 888 889 /** 890 * Signature check masks (8 bytes in total) according to the T10-PI standard: 891 * -------- -------- ------------ 892 * | GUARD | APPTAG | REFTAG | 893 * | 2B | 2B | 4B | 894 * -------- -------- ------------ 895 */ 896 enum { 897 IB_SIG_CHECK_GUARD = 0xc0, 898 IB_SIG_CHECK_APPTAG = 0x30, 899 IB_SIG_CHECK_REFTAG = 0x0f, 900 }; 901 902 /** 903 * struct ib_sig_err - signature error descriptor 904 */ 905 struct ib_sig_err { 906 enum ib_sig_err_type err_type; 907 u32 expected; 908 u32 actual; 909 u64 sig_err_offset; 910 u32 key; 911 }; 912 913 enum ib_mr_status_check { 914 IB_MR_CHECK_SIG_STATUS = 1, 915 }; 916 917 /** 918 * struct ib_mr_status - Memory region status container 919 * 920 * @fail_status: Bitmask of MR checks status. For each 921 * failed check a corresponding status bit is set. 922 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS 923 * failure. 924 */ 925 struct ib_mr_status { 926 u32 fail_status; 927 struct ib_sig_err sig_err; 928 }; 929 930 /** 931 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate 932 * enum. 933 * @mult: multiple to convert. 934 */ 935 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult); 936 937 enum rdma_ah_attr_type { 938 RDMA_AH_ATTR_TYPE_UNDEFINED, 939 RDMA_AH_ATTR_TYPE_IB, 940 RDMA_AH_ATTR_TYPE_ROCE, 941 RDMA_AH_ATTR_TYPE_OPA, 942 }; 943 944 struct ib_ah_attr { 945 u16 dlid; 946 u8 src_path_bits; 947 }; 948 949 struct roce_ah_attr { 950 u8 dmac[ETH_ALEN]; 951 }; 952 953 struct opa_ah_attr { 954 u32 dlid; 955 u8 src_path_bits; 956 bool make_grd; 957 }; 958 959 struct rdma_ah_attr { 960 struct ib_global_route grh; 961 u8 sl; 962 u8 static_rate; 963 u8 port_num; 964 u8 ah_flags; 965 enum rdma_ah_attr_type type; 966 union { 967 struct ib_ah_attr ib; 968 struct roce_ah_attr roce; 969 struct opa_ah_attr opa; 970 }; 971 }; 972 973 enum ib_wc_status { 974 IB_WC_SUCCESS, 975 IB_WC_LOC_LEN_ERR, 976 IB_WC_LOC_QP_OP_ERR, 977 IB_WC_LOC_EEC_OP_ERR, 978 IB_WC_LOC_PROT_ERR, 979 IB_WC_WR_FLUSH_ERR, 980 IB_WC_MW_BIND_ERR, 981 IB_WC_BAD_RESP_ERR, 982 IB_WC_LOC_ACCESS_ERR, 983 IB_WC_REM_INV_REQ_ERR, 984 IB_WC_REM_ACCESS_ERR, 985 IB_WC_REM_OP_ERR, 986 IB_WC_RETRY_EXC_ERR, 987 IB_WC_RNR_RETRY_EXC_ERR, 988 IB_WC_LOC_RDD_VIOL_ERR, 989 IB_WC_REM_INV_RD_REQ_ERR, 990 IB_WC_REM_ABORT_ERR, 991 IB_WC_INV_EECN_ERR, 992 IB_WC_INV_EEC_STATE_ERR, 993 IB_WC_FATAL_ERR, 994 IB_WC_RESP_TIMEOUT_ERR, 995 IB_WC_GENERAL_ERR 996 }; 997 998 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status); 999 1000 enum ib_wc_opcode { 1001 IB_WC_SEND, 1002 IB_WC_RDMA_WRITE, 1003 IB_WC_RDMA_READ, 1004 IB_WC_COMP_SWAP, 1005 IB_WC_FETCH_ADD, 1006 IB_WC_LSO, 1007 IB_WC_LOCAL_INV, 1008 IB_WC_REG_MR, 1009 IB_WC_MASKED_COMP_SWAP, 1010 IB_WC_MASKED_FETCH_ADD, 1011 /* 1012 * Set value of IB_WC_RECV so consumers can test if a completion is a 1013 * receive by testing (opcode & IB_WC_RECV). 1014 */ 1015 IB_WC_RECV = 1 << 7, 1016 IB_WC_RECV_RDMA_WITH_IMM 1017 }; 1018 1019 enum ib_wc_flags { 1020 IB_WC_GRH = 1, 1021 IB_WC_WITH_IMM = (1<<1), 1022 IB_WC_WITH_INVALIDATE = (1<<2), 1023 IB_WC_IP_CSUM_OK = (1<<3), 1024 IB_WC_WITH_SMAC = (1<<4), 1025 IB_WC_WITH_VLAN = (1<<5), 1026 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6), 1027 }; 1028 1029 struct ib_wc { 1030 union { 1031 u64 wr_id; 1032 struct ib_cqe *wr_cqe; 1033 }; 1034 enum ib_wc_status status; 1035 enum ib_wc_opcode opcode; 1036 u32 vendor_err; 1037 u32 byte_len; 1038 struct ib_qp *qp; 1039 union { 1040 __be32 imm_data; 1041 u32 invalidate_rkey; 1042 } ex; 1043 u32 src_qp; 1044 u32 slid; 1045 int wc_flags; 1046 u16 pkey_index; 1047 u8 sl; 1048 u8 dlid_path_bits; 1049 u8 port_num; /* valid only for DR SMPs on switches */ 1050 u8 smac[ETH_ALEN]; 1051 u16 vlan_id; 1052 u8 network_hdr_type; 1053 }; 1054 1055 enum ib_cq_notify_flags { 1056 IB_CQ_SOLICITED = 1 << 0, 1057 IB_CQ_NEXT_COMP = 1 << 1, 1058 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP, 1059 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2, 1060 }; 1061 1062 enum ib_srq_type { 1063 IB_SRQT_BASIC, 1064 IB_SRQT_XRC, 1065 IB_SRQT_TM, 1066 }; 1067 1068 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type) 1069 { 1070 return srq_type == IB_SRQT_XRC || 1071 srq_type == IB_SRQT_TM; 1072 } 1073 1074 enum ib_srq_attr_mask { 1075 IB_SRQ_MAX_WR = 1 << 0, 1076 IB_SRQ_LIMIT = 1 << 1, 1077 }; 1078 1079 struct ib_srq_attr { 1080 u32 max_wr; 1081 u32 max_sge; 1082 u32 srq_limit; 1083 }; 1084 1085 struct ib_srq_init_attr { 1086 void (*event_handler)(struct ib_event *, void *); 1087 void *srq_context; 1088 struct ib_srq_attr attr; 1089 enum ib_srq_type srq_type; 1090 1091 struct { 1092 struct ib_cq *cq; 1093 union { 1094 struct { 1095 struct ib_xrcd *xrcd; 1096 } xrc; 1097 1098 struct { 1099 u32 max_num_tags; 1100 } tag_matching; 1101 }; 1102 } ext; 1103 }; 1104 1105 struct ib_qp_cap { 1106 u32 max_send_wr; 1107 u32 max_recv_wr; 1108 u32 max_send_sge; 1109 u32 max_recv_sge; 1110 u32 max_inline_data; 1111 1112 /* 1113 * Maximum number of rdma_rw_ctx structures in flight at a time. 1114 * ib_create_qp() will calculate the right amount of neededed WRs 1115 * and MRs based on this. 1116 */ 1117 u32 max_rdma_ctxs; 1118 }; 1119 1120 enum ib_sig_type { 1121 IB_SIGNAL_ALL_WR, 1122 IB_SIGNAL_REQ_WR 1123 }; 1124 1125 enum ib_qp_type { 1126 /* 1127 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries 1128 * here (and in that order) since the MAD layer uses them as 1129 * indices into a 2-entry table. 1130 */ 1131 IB_QPT_SMI, 1132 IB_QPT_GSI, 1133 1134 IB_QPT_RC, 1135 IB_QPT_UC, 1136 IB_QPT_UD, 1137 IB_QPT_RAW_IPV6, 1138 IB_QPT_RAW_ETHERTYPE, 1139 IB_QPT_RAW_PACKET = 8, 1140 IB_QPT_XRC_INI = 9, 1141 IB_QPT_XRC_TGT, 1142 IB_QPT_MAX, 1143 IB_QPT_DRIVER = 0xFF, 1144 /* Reserve a range for qp types internal to the low level driver. 1145 * These qp types will not be visible at the IB core layer, so the 1146 * IB_QPT_MAX usages should not be affected in the core layer 1147 */ 1148 IB_QPT_RESERVED1 = 0x1000, 1149 IB_QPT_RESERVED2, 1150 IB_QPT_RESERVED3, 1151 IB_QPT_RESERVED4, 1152 IB_QPT_RESERVED5, 1153 IB_QPT_RESERVED6, 1154 IB_QPT_RESERVED7, 1155 IB_QPT_RESERVED8, 1156 IB_QPT_RESERVED9, 1157 IB_QPT_RESERVED10, 1158 }; 1159 1160 enum ib_qp_create_flags { 1161 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0, 1162 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1, 1163 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2, 1164 IB_QP_CREATE_MANAGED_SEND = 1 << 3, 1165 IB_QP_CREATE_MANAGED_RECV = 1 << 4, 1166 IB_QP_CREATE_NETIF_QP = 1 << 5, 1167 IB_QP_CREATE_SIGNATURE_EN = 1 << 6, 1168 /* FREE = 1 << 7, */ 1169 IB_QP_CREATE_SCATTER_FCS = 1 << 8, 1170 IB_QP_CREATE_CVLAN_STRIPPING = 1 << 9, 1171 IB_QP_CREATE_SOURCE_QPN = 1 << 10, 1172 IB_QP_CREATE_PCI_WRITE_END_PADDING = 1 << 11, 1173 /* reserve bits 26-31 for low level drivers' internal use */ 1174 IB_QP_CREATE_RESERVED_START = 1 << 26, 1175 IB_QP_CREATE_RESERVED_END = 1 << 31, 1176 }; 1177 1178 /* 1179 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler 1180 * callback to destroy the passed in QP. 1181 */ 1182 1183 struct ib_qp_init_attr { 1184 /* Consumer's event_handler callback must not block */ 1185 void (*event_handler)(struct ib_event *, void *); 1186 1187 void *qp_context; 1188 struct ib_cq *send_cq; 1189 struct ib_cq *recv_cq; 1190 struct ib_srq *srq; 1191 struct ib_xrcd *xrcd; /* XRC TGT QPs only */ 1192 struct ib_qp_cap cap; 1193 enum ib_sig_type sq_sig_type; 1194 enum ib_qp_type qp_type; 1195 u32 create_flags; 1196 1197 /* 1198 * Only needed for special QP types, or when using the RW API. 1199 */ 1200 u8 port_num; 1201 struct ib_rwq_ind_table *rwq_ind_tbl; 1202 u32 source_qpn; 1203 }; 1204 1205 struct ib_qp_open_attr { 1206 void (*event_handler)(struct ib_event *, void *); 1207 void *qp_context; 1208 u32 qp_num; 1209 enum ib_qp_type qp_type; 1210 }; 1211 1212 enum ib_rnr_timeout { 1213 IB_RNR_TIMER_655_36 = 0, 1214 IB_RNR_TIMER_000_01 = 1, 1215 IB_RNR_TIMER_000_02 = 2, 1216 IB_RNR_TIMER_000_03 = 3, 1217 IB_RNR_TIMER_000_04 = 4, 1218 IB_RNR_TIMER_000_06 = 5, 1219 IB_RNR_TIMER_000_08 = 6, 1220 IB_RNR_TIMER_000_12 = 7, 1221 IB_RNR_TIMER_000_16 = 8, 1222 IB_RNR_TIMER_000_24 = 9, 1223 IB_RNR_TIMER_000_32 = 10, 1224 IB_RNR_TIMER_000_48 = 11, 1225 IB_RNR_TIMER_000_64 = 12, 1226 IB_RNR_TIMER_000_96 = 13, 1227 IB_RNR_TIMER_001_28 = 14, 1228 IB_RNR_TIMER_001_92 = 15, 1229 IB_RNR_TIMER_002_56 = 16, 1230 IB_RNR_TIMER_003_84 = 17, 1231 IB_RNR_TIMER_005_12 = 18, 1232 IB_RNR_TIMER_007_68 = 19, 1233 IB_RNR_TIMER_010_24 = 20, 1234 IB_RNR_TIMER_015_36 = 21, 1235 IB_RNR_TIMER_020_48 = 22, 1236 IB_RNR_TIMER_030_72 = 23, 1237 IB_RNR_TIMER_040_96 = 24, 1238 IB_RNR_TIMER_061_44 = 25, 1239 IB_RNR_TIMER_081_92 = 26, 1240 IB_RNR_TIMER_122_88 = 27, 1241 IB_RNR_TIMER_163_84 = 28, 1242 IB_RNR_TIMER_245_76 = 29, 1243 IB_RNR_TIMER_327_68 = 30, 1244 IB_RNR_TIMER_491_52 = 31 1245 }; 1246 1247 enum ib_qp_attr_mask { 1248 IB_QP_STATE = 1, 1249 IB_QP_CUR_STATE = (1<<1), 1250 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2), 1251 IB_QP_ACCESS_FLAGS = (1<<3), 1252 IB_QP_PKEY_INDEX = (1<<4), 1253 IB_QP_PORT = (1<<5), 1254 IB_QP_QKEY = (1<<6), 1255 IB_QP_AV = (1<<7), 1256 IB_QP_PATH_MTU = (1<<8), 1257 IB_QP_TIMEOUT = (1<<9), 1258 IB_QP_RETRY_CNT = (1<<10), 1259 IB_QP_RNR_RETRY = (1<<11), 1260 IB_QP_RQ_PSN = (1<<12), 1261 IB_QP_MAX_QP_RD_ATOMIC = (1<<13), 1262 IB_QP_ALT_PATH = (1<<14), 1263 IB_QP_MIN_RNR_TIMER = (1<<15), 1264 IB_QP_SQ_PSN = (1<<16), 1265 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17), 1266 IB_QP_PATH_MIG_STATE = (1<<18), 1267 IB_QP_CAP = (1<<19), 1268 IB_QP_DEST_QPN = (1<<20), 1269 IB_QP_RESERVED1 = (1<<21), 1270 IB_QP_RESERVED2 = (1<<22), 1271 IB_QP_RESERVED3 = (1<<23), 1272 IB_QP_RESERVED4 = (1<<24), 1273 IB_QP_RATE_LIMIT = (1<<25), 1274 }; 1275 1276 enum ib_qp_state { 1277 IB_QPS_RESET, 1278 IB_QPS_INIT, 1279 IB_QPS_RTR, 1280 IB_QPS_RTS, 1281 IB_QPS_SQD, 1282 IB_QPS_SQE, 1283 IB_QPS_ERR 1284 }; 1285 1286 enum ib_mig_state { 1287 IB_MIG_MIGRATED, 1288 IB_MIG_REARM, 1289 IB_MIG_ARMED 1290 }; 1291 1292 enum ib_mw_type { 1293 IB_MW_TYPE_1 = 1, 1294 IB_MW_TYPE_2 = 2 1295 }; 1296 1297 struct ib_qp_attr { 1298 enum ib_qp_state qp_state; 1299 enum ib_qp_state cur_qp_state; 1300 enum ib_mtu path_mtu; 1301 enum ib_mig_state path_mig_state; 1302 u32 qkey; 1303 u32 rq_psn; 1304 u32 sq_psn; 1305 u32 dest_qp_num; 1306 int qp_access_flags; 1307 struct ib_qp_cap cap; 1308 struct rdma_ah_attr ah_attr; 1309 struct rdma_ah_attr alt_ah_attr; 1310 u16 pkey_index; 1311 u16 alt_pkey_index; 1312 u8 en_sqd_async_notify; 1313 u8 sq_draining; 1314 u8 max_rd_atomic; 1315 u8 max_dest_rd_atomic; 1316 u8 min_rnr_timer; 1317 u8 port_num; 1318 u8 timeout; 1319 u8 retry_cnt; 1320 u8 rnr_retry; 1321 u8 alt_port_num; 1322 u8 alt_timeout; 1323 u32 rate_limit; 1324 }; 1325 1326 enum ib_wr_opcode { 1327 /* These are shared with userspace */ 1328 IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE, 1329 IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM, 1330 IB_WR_SEND = IB_UVERBS_WR_SEND, 1331 IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM, 1332 IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ, 1333 IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP, 1334 IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD, 1335 IB_WR_LSO = IB_UVERBS_WR_TSO, 1336 IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV, 1337 IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV, 1338 IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV, 1339 IB_WR_MASKED_ATOMIC_CMP_AND_SWP = 1340 IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP, 1341 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD = 1342 IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD, 1343 1344 /* These are kernel only and can not be issued by userspace */ 1345 IB_WR_REG_MR = 0x20, 1346 IB_WR_REG_SIG_MR, 1347 1348 /* reserve values for low level drivers' internal use. 1349 * These values will not be used at all in the ib core layer. 1350 */ 1351 IB_WR_RESERVED1 = 0xf0, 1352 IB_WR_RESERVED2, 1353 IB_WR_RESERVED3, 1354 IB_WR_RESERVED4, 1355 IB_WR_RESERVED5, 1356 IB_WR_RESERVED6, 1357 IB_WR_RESERVED7, 1358 IB_WR_RESERVED8, 1359 IB_WR_RESERVED9, 1360 IB_WR_RESERVED10, 1361 }; 1362 1363 enum ib_send_flags { 1364 IB_SEND_FENCE = 1, 1365 IB_SEND_SIGNALED = (1<<1), 1366 IB_SEND_SOLICITED = (1<<2), 1367 IB_SEND_INLINE = (1<<3), 1368 IB_SEND_IP_CSUM = (1<<4), 1369 1370 /* reserve bits 26-31 for low level drivers' internal use */ 1371 IB_SEND_RESERVED_START = (1 << 26), 1372 IB_SEND_RESERVED_END = (1 << 31), 1373 }; 1374 1375 struct ib_sge { 1376 u64 addr; 1377 u32 length; 1378 u32 lkey; 1379 }; 1380 1381 struct ib_cqe { 1382 void (*done)(struct ib_cq *cq, struct ib_wc *wc); 1383 }; 1384 1385 struct ib_send_wr { 1386 struct ib_send_wr *next; 1387 union { 1388 u64 wr_id; 1389 struct ib_cqe *wr_cqe; 1390 }; 1391 struct ib_sge *sg_list; 1392 int num_sge; 1393 enum ib_wr_opcode opcode; 1394 int send_flags; 1395 union { 1396 __be32 imm_data; 1397 u32 invalidate_rkey; 1398 } ex; 1399 }; 1400 1401 struct ib_rdma_wr { 1402 struct ib_send_wr wr; 1403 u64 remote_addr; 1404 u32 rkey; 1405 }; 1406 1407 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr) 1408 { 1409 return container_of(wr, struct ib_rdma_wr, wr); 1410 } 1411 1412 struct ib_atomic_wr { 1413 struct ib_send_wr wr; 1414 u64 remote_addr; 1415 u64 compare_add; 1416 u64 swap; 1417 u64 compare_add_mask; 1418 u64 swap_mask; 1419 u32 rkey; 1420 }; 1421 1422 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr) 1423 { 1424 return container_of(wr, struct ib_atomic_wr, wr); 1425 } 1426 1427 struct ib_ud_wr { 1428 struct ib_send_wr wr; 1429 struct ib_ah *ah; 1430 void *header; 1431 int hlen; 1432 int mss; 1433 u32 remote_qpn; 1434 u32 remote_qkey; 1435 u16 pkey_index; /* valid for GSI only */ 1436 u8 port_num; /* valid for DR SMPs on switch only */ 1437 }; 1438 1439 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr) 1440 { 1441 return container_of(wr, struct ib_ud_wr, wr); 1442 } 1443 1444 struct ib_reg_wr { 1445 struct ib_send_wr wr; 1446 struct ib_mr *mr; 1447 u32 key; 1448 int access; 1449 }; 1450 1451 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr) 1452 { 1453 return container_of(wr, struct ib_reg_wr, wr); 1454 } 1455 1456 struct ib_sig_handover_wr { 1457 struct ib_send_wr wr; 1458 struct ib_sig_attrs *sig_attrs; 1459 struct ib_mr *sig_mr; 1460 int access_flags; 1461 struct ib_sge *prot; 1462 }; 1463 1464 static inline const struct ib_sig_handover_wr * 1465 sig_handover_wr(const struct ib_send_wr *wr) 1466 { 1467 return container_of(wr, struct ib_sig_handover_wr, wr); 1468 } 1469 1470 struct ib_recv_wr { 1471 struct ib_recv_wr *next; 1472 union { 1473 u64 wr_id; 1474 struct ib_cqe *wr_cqe; 1475 }; 1476 struct ib_sge *sg_list; 1477 int num_sge; 1478 }; 1479 1480 enum ib_access_flags { 1481 IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE, 1482 IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE, 1483 IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ, 1484 IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC, 1485 IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND, 1486 IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED, 1487 IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND, 1488 IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB, 1489 1490 IB_ACCESS_SUPPORTED = ((IB_ACCESS_HUGETLB << 1) - 1) 1491 }; 1492 1493 /* 1494 * XXX: these are apparently used for ->rereg_user_mr, no idea why they 1495 * are hidden here instead of a uapi header! 1496 */ 1497 enum ib_mr_rereg_flags { 1498 IB_MR_REREG_TRANS = 1, 1499 IB_MR_REREG_PD = (1<<1), 1500 IB_MR_REREG_ACCESS = (1<<2), 1501 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1) 1502 }; 1503 1504 struct ib_fmr_attr { 1505 int max_pages; 1506 int max_maps; 1507 u8 page_shift; 1508 }; 1509 1510 struct ib_umem; 1511 1512 enum rdma_remove_reason { 1513 /* 1514 * Userspace requested uobject deletion or initial try 1515 * to remove uobject via cleanup. Call could fail 1516 */ 1517 RDMA_REMOVE_DESTROY, 1518 /* Context deletion. This call should delete the actual object itself */ 1519 RDMA_REMOVE_CLOSE, 1520 /* Driver is being hot-unplugged. This call should delete the actual object itself */ 1521 RDMA_REMOVE_DRIVER_REMOVE, 1522 /* uobj is being cleaned-up before being committed */ 1523 RDMA_REMOVE_ABORT, 1524 }; 1525 1526 struct ib_rdmacg_object { 1527 #ifdef CONFIG_CGROUP_RDMA 1528 struct rdma_cgroup *cg; /* owner rdma cgroup */ 1529 #endif 1530 }; 1531 1532 struct ib_ucontext { 1533 struct ib_device *device; 1534 struct ib_uverbs_file *ufile; 1535 /* 1536 * 'closing' can be read by the driver only during a destroy callback, 1537 * it is set when we are closing the file descriptor and indicates 1538 * that mm_sem may be locked. 1539 */ 1540 bool closing; 1541 1542 bool cleanup_retryable; 1543 1544 void (*invalidate_range)(struct ib_umem_odp *umem_odp, 1545 unsigned long start, unsigned long end); 1546 struct mutex per_mm_list_lock; 1547 struct list_head per_mm_list; 1548 1549 struct ib_rdmacg_object cg_obj; 1550 /* 1551 * Implementation details of the RDMA core, don't use in drivers: 1552 */ 1553 struct rdma_restrack_entry res; 1554 }; 1555 1556 struct ib_uobject { 1557 u64 user_handle; /* handle given to us by userspace */ 1558 /* ufile & ucontext owning this object */ 1559 struct ib_uverbs_file *ufile; 1560 /* FIXME, save memory: ufile->context == context */ 1561 struct ib_ucontext *context; /* associated user context */ 1562 void *object; /* containing object */ 1563 struct list_head list; /* link to context's list */ 1564 struct ib_rdmacg_object cg_obj; /* rdmacg object */ 1565 int id; /* index into kernel idr */ 1566 struct kref ref; 1567 atomic_t usecnt; /* protects exclusive access */ 1568 struct rcu_head rcu; /* kfree_rcu() overhead */ 1569 1570 const struct uverbs_api_object *uapi_object; 1571 }; 1572 1573 struct ib_udata { 1574 const void __user *inbuf; 1575 void __user *outbuf; 1576 size_t inlen; 1577 size_t outlen; 1578 }; 1579 1580 struct ib_pd { 1581 u32 local_dma_lkey; 1582 u32 flags; 1583 struct ib_device *device; 1584 struct ib_uobject *uobject; 1585 atomic_t usecnt; /* count all resources */ 1586 1587 u32 unsafe_global_rkey; 1588 1589 /* 1590 * Implementation details of the RDMA core, don't use in drivers: 1591 */ 1592 struct ib_mr *__internal_mr; 1593 struct rdma_restrack_entry res; 1594 }; 1595 1596 struct ib_xrcd { 1597 struct ib_device *device; 1598 atomic_t usecnt; /* count all exposed resources */ 1599 struct inode *inode; 1600 1601 struct mutex tgt_qp_mutex; 1602 struct list_head tgt_qp_list; 1603 }; 1604 1605 struct ib_ah { 1606 struct ib_device *device; 1607 struct ib_pd *pd; 1608 struct ib_uobject *uobject; 1609 const struct ib_gid_attr *sgid_attr; 1610 enum rdma_ah_attr_type type; 1611 }; 1612 1613 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context); 1614 1615 enum ib_poll_context { 1616 IB_POLL_DIRECT, /* caller context, no hw completions */ 1617 IB_POLL_SOFTIRQ, /* poll from softirq context */ 1618 IB_POLL_WORKQUEUE, /* poll from workqueue */ 1619 IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */ 1620 }; 1621 1622 struct ib_cq { 1623 struct ib_device *device; 1624 struct ib_uobject *uobject; 1625 ib_comp_handler comp_handler; 1626 void (*event_handler)(struct ib_event *, void *); 1627 void *cq_context; 1628 int cqe; 1629 atomic_t usecnt; /* count number of work queues */ 1630 enum ib_poll_context poll_ctx; 1631 struct ib_wc *wc; 1632 union { 1633 struct irq_poll iop; 1634 struct work_struct work; 1635 }; 1636 struct workqueue_struct *comp_wq; 1637 /* 1638 * Implementation details of the RDMA core, don't use in drivers: 1639 */ 1640 struct rdma_restrack_entry res; 1641 }; 1642 1643 struct ib_srq { 1644 struct ib_device *device; 1645 struct ib_pd *pd; 1646 struct ib_uobject *uobject; 1647 void (*event_handler)(struct ib_event *, void *); 1648 void *srq_context; 1649 enum ib_srq_type srq_type; 1650 atomic_t usecnt; 1651 1652 struct { 1653 struct ib_cq *cq; 1654 union { 1655 struct { 1656 struct ib_xrcd *xrcd; 1657 u32 srq_num; 1658 } xrc; 1659 }; 1660 } ext; 1661 }; 1662 1663 enum ib_raw_packet_caps { 1664 /* Strip cvlan from incoming packet and report it in the matching work 1665 * completion is supported. 1666 */ 1667 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = (1 << 0), 1668 /* Scatter FCS field of an incoming packet to host memory is supported. 1669 */ 1670 IB_RAW_PACKET_CAP_SCATTER_FCS = (1 << 1), 1671 /* Checksum offloads are supported (for both send and receive). */ 1672 IB_RAW_PACKET_CAP_IP_CSUM = (1 << 2), 1673 /* When a packet is received for an RQ with no receive WQEs, the 1674 * packet processing is delayed. 1675 */ 1676 IB_RAW_PACKET_CAP_DELAY_DROP = (1 << 3), 1677 }; 1678 1679 enum ib_wq_type { 1680 IB_WQT_RQ 1681 }; 1682 1683 enum ib_wq_state { 1684 IB_WQS_RESET, 1685 IB_WQS_RDY, 1686 IB_WQS_ERR 1687 }; 1688 1689 struct ib_wq { 1690 struct ib_device *device; 1691 struct ib_uobject *uobject; 1692 void *wq_context; 1693 void (*event_handler)(struct ib_event *, void *); 1694 struct ib_pd *pd; 1695 struct ib_cq *cq; 1696 u32 wq_num; 1697 enum ib_wq_state state; 1698 enum ib_wq_type wq_type; 1699 atomic_t usecnt; 1700 }; 1701 1702 enum ib_wq_flags { 1703 IB_WQ_FLAGS_CVLAN_STRIPPING = 1 << 0, 1704 IB_WQ_FLAGS_SCATTER_FCS = 1 << 1, 1705 IB_WQ_FLAGS_DELAY_DROP = 1 << 2, 1706 IB_WQ_FLAGS_PCI_WRITE_END_PADDING = 1 << 3, 1707 }; 1708 1709 struct ib_wq_init_attr { 1710 void *wq_context; 1711 enum ib_wq_type wq_type; 1712 u32 max_wr; 1713 u32 max_sge; 1714 struct ib_cq *cq; 1715 void (*event_handler)(struct ib_event *, void *); 1716 u32 create_flags; /* Use enum ib_wq_flags */ 1717 }; 1718 1719 enum ib_wq_attr_mask { 1720 IB_WQ_STATE = 1 << 0, 1721 IB_WQ_CUR_STATE = 1 << 1, 1722 IB_WQ_FLAGS = 1 << 2, 1723 }; 1724 1725 struct ib_wq_attr { 1726 enum ib_wq_state wq_state; 1727 enum ib_wq_state curr_wq_state; 1728 u32 flags; /* Use enum ib_wq_flags */ 1729 u32 flags_mask; /* Use enum ib_wq_flags */ 1730 }; 1731 1732 struct ib_rwq_ind_table { 1733 struct ib_device *device; 1734 struct ib_uobject *uobject; 1735 atomic_t usecnt; 1736 u32 ind_tbl_num; 1737 u32 log_ind_tbl_size; 1738 struct ib_wq **ind_tbl; 1739 }; 1740 1741 struct ib_rwq_ind_table_init_attr { 1742 u32 log_ind_tbl_size; 1743 /* Each entry is a pointer to Receive Work Queue */ 1744 struct ib_wq **ind_tbl; 1745 }; 1746 1747 enum port_pkey_state { 1748 IB_PORT_PKEY_NOT_VALID = 0, 1749 IB_PORT_PKEY_VALID = 1, 1750 IB_PORT_PKEY_LISTED = 2, 1751 }; 1752 1753 struct ib_qp_security; 1754 1755 struct ib_port_pkey { 1756 enum port_pkey_state state; 1757 u16 pkey_index; 1758 u8 port_num; 1759 struct list_head qp_list; 1760 struct list_head to_error_list; 1761 struct ib_qp_security *sec; 1762 }; 1763 1764 struct ib_ports_pkeys { 1765 struct ib_port_pkey main; 1766 struct ib_port_pkey alt; 1767 }; 1768 1769 struct ib_qp_security { 1770 struct ib_qp *qp; 1771 struct ib_device *dev; 1772 /* Hold this mutex when changing port and pkey settings. */ 1773 struct mutex mutex; 1774 struct ib_ports_pkeys *ports_pkeys; 1775 /* A list of all open shared QP handles. Required to enforce security 1776 * properly for all users of a shared QP. 1777 */ 1778 struct list_head shared_qp_list; 1779 void *security; 1780 bool destroying; 1781 atomic_t error_list_count; 1782 struct completion error_complete; 1783 int error_comps_pending; 1784 }; 1785 1786 /* 1787 * @max_write_sge: Maximum SGE elements per RDMA WRITE request. 1788 * @max_read_sge: Maximum SGE elements per RDMA READ request. 1789 */ 1790 struct ib_qp { 1791 struct ib_device *device; 1792 struct ib_pd *pd; 1793 struct ib_cq *send_cq; 1794 struct ib_cq *recv_cq; 1795 spinlock_t mr_lock; 1796 int mrs_used; 1797 struct list_head rdma_mrs; 1798 struct list_head sig_mrs; 1799 struct ib_srq *srq; 1800 struct ib_xrcd *xrcd; /* XRC TGT QPs only */ 1801 struct list_head xrcd_list; 1802 1803 /* count times opened, mcast attaches, flow attaches */ 1804 atomic_t usecnt; 1805 struct list_head open_list; 1806 struct ib_qp *real_qp; 1807 struct ib_uobject *uobject; 1808 void (*event_handler)(struct ib_event *, void *); 1809 void *qp_context; 1810 /* sgid_attrs associated with the AV's */ 1811 const struct ib_gid_attr *av_sgid_attr; 1812 const struct ib_gid_attr *alt_path_sgid_attr; 1813 u32 qp_num; 1814 u32 max_write_sge; 1815 u32 max_read_sge; 1816 enum ib_qp_type qp_type; 1817 struct ib_rwq_ind_table *rwq_ind_tbl; 1818 struct ib_qp_security *qp_sec; 1819 u8 port; 1820 1821 /* 1822 * Implementation details of the RDMA core, don't use in drivers: 1823 */ 1824 struct rdma_restrack_entry res; 1825 }; 1826 1827 struct ib_dm { 1828 struct ib_device *device; 1829 u32 length; 1830 u32 flags; 1831 struct ib_uobject *uobject; 1832 atomic_t usecnt; 1833 }; 1834 1835 struct ib_mr { 1836 struct ib_device *device; 1837 struct ib_pd *pd; 1838 u32 lkey; 1839 u32 rkey; 1840 u64 iova; 1841 u64 length; 1842 unsigned int page_size; 1843 bool need_inval; 1844 union { 1845 struct ib_uobject *uobject; /* user */ 1846 struct list_head qp_entry; /* FR */ 1847 }; 1848 1849 struct ib_dm *dm; 1850 1851 /* 1852 * Implementation details of the RDMA core, don't use in drivers: 1853 */ 1854 struct rdma_restrack_entry res; 1855 }; 1856 1857 struct ib_mw { 1858 struct ib_device *device; 1859 struct ib_pd *pd; 1860 struct ib_uobject *uobject; 1861 u32 rkey; 1862 enum ib_mw_type type; 1863 }; 1864 1865 struct ib_fmr { 1866 struct ib_device *device; 1867 struct ib_pd *pd; 1868 struct list_head list; 1869 u32 lkey; 1870 u32 rkey; 1871 }; 1872 1873 /* Supported steering options */ 1874 enum ib_flow_attr_type { 1875 /* steering according to rule specifications */ 1876 IB_FLOW_ATTR_NORMAL = 0x0, 1877 /* default unicast and multicast rule - 1878 * receive all Eth traffic which isn't steered to any QP 1879 */ 1880 IB_FLOW_ATTR_ALL_DEFAULT = 0x1, 1881 /* default multicast rule - 1882 * receive all Eth multicast traffic which isn't steered to any QP 1883 */ 1884 IB_FLOW_ATTR_MC_DEFAULT = 0x2, 1885 /* sniffer rule - receive all port traffic */ 1886 IB_FLOW_ATTR_SNIFFER = 0x3 1887 }; 1888 1889 /* Supported steering header types */ 1890 enum ib_flow_spec_type { 1891 /* L2 headers*/ 1892 IB_FLOW_SPEC_ETH = 0x20, 1893 IB_FLOW_SPEC_IB = 0x22, 1894 /* L3 header*/ 1895 IB_FLOW_SPEC_IPV4 = 0x30, 1896 IB_FLOW_SPEC_IPV6 = 0x31, 1897 IB_FLOW_SPEC_ESP = 0x34, 1898 /* L4 headers*/ 1899 IB_FLOW_SPEC_TCP = 0x40, 1900 IB_FLOW_SPEC_UDP = 0x41, 1901 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50, 1902 IB_FLOW_SPEC_GRE = 0x51, 1903 IB_FLOW_SPEC_MPLS = 0x60, 1904 IB_FLOW_SPEC_INNER = 0x100, 1905 /* Actions */ 1906 IB_FLOW_SPEC_ACTION_TAG = 0x1000, 1907 IB_FLOW_SPEC_ACTION_DROP = 0x1001, 1908 IB_FLOW_SPEC_ACTION_HANDLE = 0x1002, 1909 IB_FLOW_SPEC_ACTION_COUNT = 0x1003, 1910 }; 1911 #define IB_FLOW_SPEC_LAYER_MASK 0xF0 1912 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10 1913 1914 /* Flow steering rule priority is set according to it's domain. 1915 * Lower domain value means higher priority. 1916 */ 1917 enum ib_flow_domain { 1918 IB_FLOW_DOMAIN_USER, 1919 IB_FLOW_DOMAIN_ETHTOOL, 1920 IB_FLOW_DOMAIN_RFS, 1921 IB_FLOW_DOMAIN_NIC, 1922 IB_FLOW_DOMAIN_NUM /* Must be last */ 1923 }; 1924 1925 enum ib_flow_flags { 1926 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */ 1927 IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */ 1928 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 3 /* Must be last */ 1929 }; 1930 1931 struct ib_flow_eth_filter { 1932 u8 dst_mac[6]; 1933 u8 src_mac[6]; 1934 __be16 ether_type; 1935 __be16 vlan_tag; 1936 /* Must be last */ 1937 u8 real_sz[0]; 1938 }; 1939 1940 struct ib_flow_spec_eth { 1941 u32 type; 1942 u16 size; 1943 struct ib_flow_eth_filter val; 1944 struct ib_flow_eth_filter mask; 1945 }; 1946 1947 struct ib_flow_ib_filter { 1948 __be16 dlid; 1949 __u8 sl; 1950 /* Must be last */ 1951 u8 real_sz[0]; 1952 }; 1953 1954 struct ib_flow_spec_ib { 1955 u32 type; 1956 u16 size; 1957 struct ib_flow_ib_filter val; 1958 struct ib_flow_ib_filter mask; 1959 }; 1960 1961 /* IPv4 header flags */ 1962 enum ib_ipv4_flags { 1963 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */ 1964 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the 1965 last have this flag set */ 1966 }; 1967 1968 struct ib_flow_ipv4_filter { 1969 __be32 src_ip; 1970 __be32 dst_ip; 1971 u8 proto; 1972 u8 tos; 1973 u8 ttl; 1974 u8 flags; 1975 /* Must be last */ 1976 u8 real_sz[0]; 1977 }; 1978 1979 struct ib_flow_spec_ipv4 { 1980 u32 type; 1981 u16 size; 1982 struct ib_flow_ipv4_filter val; 1983 struct ib_flow_ipv4_filter mask; 1984 }; 1985 1986 struct ib_flow_ipv6_filter { 1987 u8 src_ip[16]; 1988 u8 dst_ip[16]; 1989 __be32 flow_label; 1990 u8 next_hdr; 1991 u8 traffic_class; 1992 u8 hop_limit; 1993 /* Must be last */ 1994 u8 real_sz[0]; 1995 }; 1996 1997 struct ib_flow_spec_ipv6 { 1998 u32 type; 1999 u16 size; 2000 struct ib_flow_ipv6_filter val; 2001 struct ib_flow_ipv6_filter mask; 2002 }; 2003 2004 struct ib_flow_tcp_udp_filter { 2005 __be16 dst_port; 2006 __be16 src_port; 2007 /* Must be last */ 2008 u8 real_sz[0]; 2009 }; 2010 2011 struct ib_flow_spec_tcp_udp { 2012 u32 type; 2013 u16 size; 2014 struct ib_flow_tcp_udp_filter val; 2015 struct ib_flow_tcp_udp_filter mask; 2016 }; 2017 2018 struct ib_flow_tunnel_filter { 2019 __be32 tunnel_id; 2020 u8 real_sz[0]; 2021 }; 2022 2023 /* ib_flow_spec_tunnel describes the Vxlan tunnel 2024 * the tunnel_id from val has the vni value 2025 */ 2026 struct ib_flow_spec_tunnel { 2027 u32 type; 2028 u16 size; 2029 struct ib_flow_tunnel_filter val; 2030 struct ib_flow_tunnel_filter mask; 2031 }; 2032 2033 struct ib_flow_esp_filter { 2034 __be32 spi; 2035 __be32 seq; 2036 /* Must be last */ 2037 u8 real_sz[0]; 2038 }; 2039 2040 struct ib_flow_spec_esp { 2041 u32 type; 2042 u16 size; 2043 struct ib_flow_esp_filter val; 2044 struct ib_flow_esp_filter mask; 2045 }; 2046 2047 struct ib_flow_gre_filter { 2048 __be16 c_ks_res0_ver; 2049 __be16 protocol; 2050 __be32 key; 2051 /* Must be last */ 2052 u8 real_sz[0]; 2053 }; 2054 2055 struct ib_flow_spec_gre { 2056 u32 type; 2057 u16 size; 2058 struct ib_flow_gre_filter val; 2059 struct ib_flow_gre_filter mask; 2060 }; 2061 2062 struct ib_flow_mpls_filter { 2063 __be32 tag; 2064 /* Must be last */ 2065 u8 real_sz[0]; 2066 }; 2067 2068 struct ib_flow_spec_mpls { 2069 u32 type; 2070 u16 size; 2071 struct ib_flow_mpls_filter val; 2072 struct ib_flow_mpls_filter mask; 2073 }; 2074 2075 struct ib_flow_spec_action_tag { 2076 enum ib_flow_spec_type type; 2077 u16 size; 2078 u32 tag_id; 2079 }; 2080 2081 struct ib_flow_spec_action_drop { 2082 enum ib_flow_spec_type type; 2083 u16 size; 2084 }; 2085 2086 struct ib_flow_spec_action_handle { 2087 enum ib_flow_spec_type type; 2088 u16 size; 2089 struct ib_flow_action *act; 2090 }; 2091 2092 enum ib_counters_description { 2093 IB_COUNTER_PACKETS, 2094 IB_COUNTER_BYTES, 2095 }; 2096 2097 struct ib_flow_spec_action_count { 2098 enum ib_flow_spec_type type; 2099 u16 size; 2100 struct ib_counters *counters; 2101 }; 2102 2103 union ib_flow_spec { 2104 struct { 2105 u32 type; 2106 u16 size; 2107 }; 2108 struct ib_flow_spec_eth eth; 2109 struct ib_flow_spec_ib ib; 2110 struct ib_flow_spec_ipv4 ipv4; 2111 struct ib_flow_spec_tcp_udp tcp_udp; 2112 struct ib_flow_spec_ipv6 ipv6; 2113 struct ib_flow_spec_tunnel tunnel; 2114 struct ib_flow_spec_esp esp; 2115 struct ib_flow_spec_gre gre; 2116 struct ib_flow_spec_mpls mpls; 2117 struct ib_flow_spec_action_tag flow_tag; 2118 struct ib_flow_spec_action_drop drop; 2119 struct ib_flow_spec_action_handle action; 2120 struct ib_flow_spec_action_count flow_count; 2121 }; 2122 2123 struct ib_flow_attr { 2124 enum ib_flow_attr_type type; 2125 u16 size; 2126 u16 priority; 2127 u32 flags; 2128 u8 num_of_specs; 2129 u8 port; 2130 union ib_flow_spec flows[]; 2131 }; 2132 2133 struct ib_flow { 2134 struct ib_qp *qp; 2135 struct ib_device *device; 2136 struct ib_uobject *uobject; 2137 }; 2138 2139 enum ib_flow_action_type { 2140 IB_FLOW_ACTION_UNSPECIFIED, 2141 IB_FLOW_ACTION_ESP = 1, 2142 }; 2143 2144 struct ib_flow_action_attrs_esp_keymats { 2145 enum ib_uverbs_flow_action_esp_keymat protocol; 2146 union { 2147 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm; 2148 } keymat; 2149 }; 2150 2151 struct ib_flow_action_attrs_esp_replays { 2152 enum ib_uverbs_flow_action_esp_replay protocol; 2153 union { 2154 struct ib_uverbs_flow_action_esp_replay_bmp bmp; 2155 } replay; 2156 }; 2157 2158 enum ib_flow_action_attrs_esp_flags { 2159 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags 2160 * This is done in order to share the same flags between user-space and 2161 * kernel and spare an unnecessary translation. 2162 */ 2163 2164 /* Kernel flags */ 2165 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32, 2166 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33, 2167 }; 2168 2169 struct ib_flow_spec_list { 2170 struct ib_flow_spec_list *next; 2171 union ib_flow_spec spec; 2172 }; 2173 2174 struct ib_flow_action_attrs_esp { 2175 struct ib_flow_action_attrs_esp_keymats *keymat; 2176 struct ib_flow_action_attrs_esp_replays *replay; 2177 struct ib_flow_spec_list *encap; 2178 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled. 2179 * Value of 0 is a valid value. 2180 */ 2181 u32 esn; 2182 u32 spi; 2183 u32 seq; 2184 u32 tfc_pad; 2185 /* Use enum ib_flow_action_attrs_esp_flags */ 2186 u64 flags; 2187 u64 hard_limit_pkts; 2188 }; 2189 2190 struct ib_flow_action { 2191 struct ib_device *device; 2192 struct ib_uobject *uobject; 2193 enum ib_flow_action_type type; 2194 atomic_t usecnt; 2195 }; 2196 2197 struct ib_mad_hdr; 2198 struct ib_grh; 2199 2200 enum ib_process_mad_flags { 2201 IB_MAD_IGNORE_MKEY = 1, 2202 IB_MAD_IGNORE_BKEY = 2, 2203 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY 2204 }; 2205 2206 enum ib_mad_result { 2207 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */ 2208 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */ 2209 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */ 2210 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */ 2211 }; 2212 2213 struct ib_port_cache { 2214 u64 subnet_prefix; 2215 struct ib_pkey_cache *pkey; 2216 struct ib_gid_table *gid; 2217 u8 lmc; 2218 enum ib_port_state port_state; 2219 }; 2220 2221 struct ib_cache { 2222 rwlock_t lock; 2223 struct ib_event_handler event_handler; 2224 }; 2225 2226 struct ib_port_immutable { 2227 int pkey_tbl_len; 2228 int gid_tbl_len; 2229 u32 core_cap_flags; 2230 u32 max_mad_size; 2231 }; 2232 2233 struct ib_port_data { 2234 struct ib_device *ib_dev; 2235 2236 struct ib_port_immutable immutable; 2237 2238 spinlock_t pkey_list_lock; 2239 struct list_head pkey_list; 2240 2241 struct ib_port_cache cache; 2242 2243 spinlock_t netdev_lock; 2244 struct net_device __rcu *netdev; 2245 struct hlist_node ndev_hash_link; 2246 }; 2247 2248 /* rdma netdev type - specifies protocol type */ 2249 enum rdma_netdev_t { 2250 RDMA_NETDEV_OPA_VNIC, 2251 RDMA_NETDEV_IPOIB, 2252 }; 2253 2254 /** 2255 * struct rdma_netdev - rdma netdev 2256 * For cases where netstack interfacing is required. 2257 */ 2258 struct rdma_netdev { 2259 void *clnt_priv; 2260 struct ib_device *hca; 2261 u8 port_num; 2262 2263 /* 2264 * cleanup function must be specified. 2265 * FIXME: This is only used for OPA_VNIC and that usage should be 2266 * removed too. 2267 */ 2268 void (*free_rdma_netdev)(struct net_device *netdev); 2269 2270 /* control functions */ 2271 void (*set_id)(struct net_device *netdev, int id); 2272 /* send packet */ 2273 int (*send)(struct net_device *dev, struct sk_buff *skb, 2274 struct ib_ah *address, u32 dqpn); 2275 /* multicast */ 2276 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca, 2277 union ib_gid *gid, u16 mlid, 2278 int set_qkey, u32 qkey); 2279 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca, 2280 union ib_gid *gid, u16 mlid); 2281 }; 2282 2283 struct rdma_netdev_alloc_params { 2284 size_t sizeof_priv; 2285 unsigned int txqs; 2286 unsigned int rxqs; 2287 void *param; 2288 2289 int (*initialize_rdma_netdev)(struct ib_device *device, u8 port_num, 2290 struct net_device *netdev, void *param); 2291 }; 2292 2293 struct ib_counters { 2294 struct ib_device *device; 2295 struct ib_uobject *uobject; 2296 /* num of objects attached */ 2297 atomic_t usecnt; 2298 }; 2299 2300 struct ib_counters_read_attr { 2301 u64 *counters_buff; 2302 u32 ncounters; 2303 u32 flags; /* use enum ib_read_counters_flags */ 2304 }; 2305 2306 struct uverbs_attr_bundle; 2307 struct iw_cm_id; 2308 struct iw_cm_conn_param; 2309 2310 #define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member) \ 2311 .size_##ib_struct = \ 2312 (sizeof(struct drv_struct) + \ 2313 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) + \ 2314 BUILD_BUG_ON_ZERO( \ 2315 !__same_type(((struct drv_struct *)NULL)->member, \ 2316 struct ib_struct))) 2317 2318 #define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp) \ 2319 ((struct ib_type *)kzalloc(ib_dev->ops.size_##ib_type, gfp)) 2320 2321 #define rdma_zalloc_drv_obj(ib_dev, ib_type) \ 2322 rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL) 2323 2324 #define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct 2325 2326 /** 2327 * struct ib_device_ops - InfiniBand device operations 2328 * This structure defines all the InfiniBand device operations, providers will 2329 * need to define the supported operations, otherwise they will be set to null. 2330 */ 2331 struct ib_device_ops { 2332 int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr, 2333 const struct ib_send_wr **bad_send_wr); 2334 int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr, 2335 const struct ib_recv_wr **bad_recv_wr); 2336 void (*drain_rq)(struct ib_qp *qp); 2337 void (*drain_sq)(struct ib_qp *qp); 2338 int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc); 2339 int (*peek_cq)(struct ib_cq *cq, int wc_cnt); 2340 int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags); 2341 int (*req_ncomp_notif)(struct ib_cq *cq, int wc_cnt); 2342 int (*post_srq_recv)(struct ib_srq *srq, 2343 const struct ib_recv_wr *recv_wr, 2344 const struct ib_recv_wr **bad_recv_wr); 2345 int (*process_mad)(struct ib_device *device, int process_mad_flags, 2346 u8 port_num, const struct ib_wc *in_wc, 2347 const struct ib_grh *in_grh, 2348 const struct ib_mad_hdr *in_mad, size_t in_mad_size, 2349 struct ib_mad_hdr *out_mad, size_t *out_mad_size, 2350 u16 *out_mad_pkey_index); 2351 int (*query_device)(struct ib_device *device, 2352 struct ib_device_attr *device_attr, 2353 struct ib_udata *udata); 2354 int (*modify_device)(struct ib_device *device, int device_modify_mask, 2355 struct ib_device_modify *device_modify); 2356 void (*get_dev_fw_str)(struct ib_device *device, char *str); 2357 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev, 2358 int comp_vector); 2359 int (*query_port)(struct ib_device *device, u8 port_num, 2360 struct ib_port_attr *port_attr); 2361 int (*modify_port)(struct ib_device *device, u8 port_num, 2362 int port_modify_mask, 2363 struct ib_port_modify *port_modify); 2364 /** 2365 * The following mandatory functions are used only at device 2366 * registration. Keep functions such as these at the end of this 2367 * structure to avoid cache line misses when accessing struct ib_device 2368 * in fast paths. 2369 */ 2370 int (*get_port_immutable)(struct ib_device *device, u8 port_num, 2371 struct ib_port_immutable *immutable); 2372 enum rdma_link_layer (*get_link_layer)(struct ib_device *device, 2373 u8 port_num); 2374 /** 2375 * When calling get_netdev, the HW vendor's driver should return the 2376 * net device of device @device at port @port_num or NULL if such 2377 * a net device doesn't exist. The vendor driver should call dev_hold 2378 * on this net device. The HW vendor's device driver must guarantee 2379 * that this function returns NULL before the net device has finished 2380 * NETDEV_UNREGISTER state. 2381 */ 2382 struct net_device *(*get_netdev)(struct ib_device *device, u8 port_num); 2383 /** 2384 * rdma netdev operation 2385 * 2386 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params 2387 * must return -EOPNOTSUPP if it doesn't support the specified type. 2388 */ 2389 struct net_device *(*alloc_rdma_netdev)( 2390 struct ib_device *device, u8 port_num, enum rdma_netdev_t type, 2391 const char *name, unsigned char name_assign_type, 2392 void (*setup)(struct net_device *)); 2393 2394 int (*rdma_netdev_get_params)(struct ib_device *device, u8 port_num, 2395 enum rdma_netdev_t type, 2396 struct rdma_netdev_alloc_params *params); 2397 /** 2398 * query_gid should be return GID value for @device, when @port_num 2399 * link layer is either IB or iWarp. It is no-op if @port_num port 2400 * is RoCE link layer. 2401 */ 2402 int (*query_gid)(struct ib_device *device, u8 port_num, int index, 2403 union ib_gid *gid); 2404 /** 2405 * When calling add_gid, the HW vendor's driver should add the gid 2406 * of device of port at gid index available at @attr. Meta-info of 2407 * that gid (for example, the network device related to this gid) is 2408 * available at @attr. @context allows the HW vendor driver to store 2409 * extra information together with a GID entry. The HW vendor driver may 2410 * allocate memory to contain this information and store it in @context 2411 * when a new GID entry is written to. Params are consistent until the 2412 * next call of add_gid or delete_gid. The function should return 0 on 2413 * success or error otherwise. The function could be called 2414 * concurrently for different ports. This function is only called when 2415 * roce_gid_table is used. 2416 */ 2417 int (*add_gid)(const struct ib_gid_attr *attr, void **context); 2418 /** 2419 * When calling del_gid, the HW vendor's driver should delete the 2420 * gid of device @device at gid index gid_index of port port_num 2421 * available in @attr. 2422 * Upon the deletion of a GID entry, the HW vendor must free any 2423 * allocated memory. The caller will clear @context afterwards. 2424 * This function is only called when roce_gid_table is used. 2425 */ 2426 int (*del_gid)(const struct ib_gid_attr *attr, void **context); 2427 int (*query_pkey)(struct ib_device *device, u8 port_num, u16 index, 2428 u16 *pkey); 2429 int (*alloc_ucontext)(struct ib_ucontext *context, 2430 struct ib_udata *udata); 2431 void (*dealloc_ucontext)(struct ib_ucontext *context); 2432 int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma); 2433 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext); 2434 int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata); 2435 void (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata); 2436 int (*create_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr, 2437 u32 flags, struct ib_udata *udata); 2438 int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr); 2439 int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr); 2440 void (*destroy_ah)(struct ib_ah *ah, u32 flags); 2441 int (*create_srq)(struct ib_srq *srq, 2442 struct ib_srq_init_attr *srq_init_attr, 2443 struct ib_udata *udata); 2444 int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr, 2445 enum ib_srq_attr_mask srq_attr_mask, 2446 struct ib_udata *udata); 2447 int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr); 2448 void (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata); 2449 struct ib_qp *(*create_qp)(struct ib_pd *pd, 2450 struct ib_qp_init_attr *qp_init_attr, 2451 struct ib_udata *udata); 2452 int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr, 2453 int qp_attr_mask, struct ib_udata *udata); 2454 int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr, 2455 int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr); 2456 int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata); 2457 struct ib_cq *(*create_cq)(struct ib_device *device, 2458 const struct ib_cq_init_attr *attr, 2459 struct ib_udata *udata); 2460 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period); 2461 int (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata); 2462 int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata); 2463 struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags); 2464 struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length, 2465 u64 virt_addr, int mr_access_flags, 2466 struct ib_udata *udata); 2467 int (*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start, u64 length, 2468 u64 virt_addr, int mr_access_flags, 2469 struct ib_pd *pd, struct ib_udata *udata); 2470 int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata); 2471 struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type, 2472 u32 max_num_sg, struct ib_udata *udata); 2473 int (*advise_mr)(struct ib_pd *pd, 2474 enum ib_uverbs_advise_mr_advice advice, u32 flags, 2475 struct ib_sge *sg_list, u32 num_sge, 2476 struct uverbs_attr_bundle *attrs); 2477 int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents, 2478 unsigned int *sg_offset); 2479 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask, 2480 struct ib_mr_status *mr_status); 2481 struct ib_mw *(*alloc_mw)(struct ib_pd *pd, enum ib_mw_type type, 2482 struct ib_udata *udata); 2483 int (*dealloc_mw)(struct ib_mw *mw); 2484 struct ib_fmr *(*alloc_fmr)(struct ib_pd *pd, int mr_access_flags, 2485 struct ib_fmr_attr *fmr_attr); 2486 int (*map_phys_fmr)(struct ib_fmr *fmr, u64 *page_list, int list_len, 2487 u64 iova); 2488 int (*unmap_fmr)(struct list_head *fmr_list); 2489 int (*dealloc_fmr)(struct ib_fmr *fmr); 2490 int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid); 2491 int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid); 2492 struct ib_xrcd *(*alloc_xrcd)(struct ib_device *device, 2493 struct ib_udata *udata); 2494 int (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata); 2495 struct ib_flow *(*create_flow)(struct ib_qp *qp, 2496 struct ib_flow_attr *flow_attr, 2497 int domain, struct ib_udata *udata); 2498 int (*destroy_flow)(struct ib_flow *flow_id); 2499 struct ib_flow_action *(*create_flow_action_esp)( 2500 struct ib_device *device, 2501 const struct ib_flow_action_attrs_esp *attr, 2502 struct uverbs_attr_bundle *attrs); 2503 int (*destroy_flow_action)(struct ib_flow_action *action); 2504 int (*modify_flow_action_esp)( 2505 struct ib_flow_action *action, 2506 const struct ib_flow_action_attrs_esp *attr, 2507 struct uverbs_attr_bundle *attrs); 2508 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port, 2509 int state); 2510 int (*get_vf_config)(struct ib_device *device, int vf, u8 port, 2511 struct ifla_vf_info *ivf); 2512 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port, 2513 struct ifla_vf_stats *stats); 2514 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid, 2515 int type); 2516 struct ib_wq *(*create_wq)(struct ib_pd *pd, 2517 struct ib_wq_init_attr *init_attr, 2518 struct ib_udata *udata); 2519 int (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata); 2520 int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr, 2521 u32 wq_attr_mask, struct ib_udata *udata); 2522 struct ib_rwq_ind_table *(*create_rwq_ind_table)( 2523 struct ib_device *device, 2524 struct ib_rwq_ind_table_init_attr *init_attr, 2525 struct ib_udata *udata); 2526 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table); 2527 struct ib_dm *(*alloc_dm)(struct ib_device *device, 2528 struct ib_ucontext *context, 2529 struct ib_dm_alloc_attr *attr, 2530 struct uverbs_attr_bundle *attrs); 2531 int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs); 2532 struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm, 2533 struct ib_dm_mr_attr *attr, 2534 struct uverbs_attr_bundle *attrs); 2535 struct ib_counters *(*create_counters)( 2536 struct ib_device *device, struct uverbs_attr_bundle *attrs); 2537 int (*destroy_counters)(struct ib_counters *counters); 2538 int (*read_counters)(struct ib_counters *counters, 2539 struct ib_counters_read_attr *counters_read_attr, 2540 struct uverbs_attr_bundle *attrs); 2541 /** 2542 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the 2543 * driver initialized data. The struct is kfree()'ed by the sysfs 2544 * core when the device is removed. A lifespan of -1 in the return 2545 * struct tells the core to set a default lifespan. 2546 */ 2547 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device, 2548 u8 port_num); 2549 /** 2550 * get_hw_stats - Fill in the counter value(s) in the stats struct. 2551 * @index - The index in the value array we wish to have updated, or 2552 * num_counters if we want all stats updated 2553 * Return codes - 2554 * < 0 - Error, no counters updated 2555 * index - Updated the single counter pointed to by index 2556 * num_counters - Updated all counters (will reset the timestamp 2557 * and prevent further calls for lifespan milliseconds) 2558 * Drivers are allowed to update all counters in leiu of just the 2559 * one given in index at their option 2560 */ 2561 int (*get_hw_stats)(struct ib_device *device, 2562 struct rdma_hw_stats *stats, u8 port, int index); 2563 /* 2564 * This function is called once for each port when a ib device is 2565 * registered. 2566 */ 2567 int (*init_port)(struct ib_device *device, u8 port_num, 2568 struct kobject *port_sysfs); 2569 /** 2570 * Allows rdma drivers to add their own restrack attributes. 2571 */ 2572 int (*fill_res_entry)(struct sk_buff *msg, 2573 struct rdma_restrack_entry *entry); 2574 2575 /* Device lifecycle callbacks */ 2576 /* 2577 * Called after the device becomes registered, before clients are 2578 * attached 2579 */ 2580 int (*enable_driver)(struct ib_device *dev); 2581 /* 2582 * This is called as part of ib_dealloc_device(). 2583 */ 2584 void (*dealloc_driver)(struct ib_device *dev); 2585 2586 /* iWarp CM callbacks */ 2587 void (*iw_add_ref)(struct ib_qp *qp); 2588 void (*iw_rem_ref)(struct ib_qp *qp); 2589 struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn); 2590 int (*iw_connect)(struct iw_cm_id *cm_id, 2591 struct iw_cm_conn_param *conn_param); 2592 int (*iw_accept)(struct iw_cm_id *cm_id, 2593 struct iw_cm_conn_param *conn_param); 2594 int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata, 2595 u8 pdata_len); 2596 int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog); 2597 int (*iw_destroy_listen)(struct iw_cm_id *cm_id); 2598 2599 DECLARE_RDMA_OBJ_SIZE(ib_ah); 2600 DECLARE_RDMA_OBJ_SIZE(ib_pd); 2601 DECLARE_RDMA_OBJ_SIZE(ib_srq); 2602 DECLARE_RDMA_OBJ_SIZE(ib_ucontext); 2603 }; 2604 2605 struct ib_core_device { 2606 /* device must be the first element in structure until, 2607 * union of ib_core_device and device exists in ib_device. 2608 */ 2609 struct device dev; 2610 possible_net_t rdma_net; 2611 struct kobject *ports_kobj; 2612 struct list_head port_list; 2613 struct ib_device *owner; /* reach back to owner ib_device */ 2614 }; 2615 2616 struct rdma_restrack_root; 2617 struct ib_device { 2618 /* Do not access @dma_device directly from ULP nor from HW drivers. */ 2619 struct device *dma_device; 2620 struct ib_device_ops ops; 2621 char name[IB_DEVICE_NAME_MAX]; 2622 struct rcu_head rcu_head; 2623 2624 struct list_head event_handler_list; 2625 spinlock_t event_handler_lock; 2626 2627 struct rw_semaphore client_data_rwsem; 2628 struct xarray client_data; 2629 struct mutex unregistration_lock; 2630 2631 struct ib_cache cache; 2632 /** 2633 * port_data is indexed by port number 2634 */ 2635 struct ib_port_data *port_data; 2636 2637 int num_comp_vectors; 2638 2639 struct module *owner; 2640 union { 2641 struct device dev; 2642 struct ib_core_device coredev; 2643 }; 2644 2645 /* First group for device attributes, 2646 * Second group for driver provided attributes (optional). 2647 * It is NULL terminated array. 2648 */ 2649 const struct attribute_group *groups[3]; 2650 2651 int uverbs_abi_ver; 2652 u64 uverbs_cmd_mask; 2653 u64 uverbs_ex_cmd_mask; 2654 2655 char node_desc[IB_DEVICE_NODE_DESC_MAX]; 2656 __be64 node_guid; 2657 u32 local_dma_lkey; 2658 u16 is_switch:1; 2659 /* Indicates kernel verbs support, should not be used in drivers */ 2660 u16 kverbs_provider:1; 2661 u8 node_type; 2662 u8 phys_port_cnt; 2663 struct ib_device_attr attrs; 2664 struct attribute_group *hw_stats_ag; 2665 struct rdma_hw_stats *hw_stats; 2666 2667 #ifdef CONFIG_CGROUP_RDMA 2668 struct rdmacg_device cg_device; 2669 #endif 2670 2671 u32 index; 2672 struct rdma_restrack_root *res; 2673 2674 const struct uapi_definition *driver_def; 2675 enum rdma_driver_id driver_id; 2676 2677 /* 2678 * Positive refcount indicates that the device is currently 2679 * registered and cannot be unregistered. 2680 */ 2681 refcount_t refcount; 2682 struct completion unreg_completion; 2683 struct work_struct unregistration_work; 2684 2685 const struct rdma_link_ops *link_ops; 2686 2687 /* Protects compat_devs xarray modifications */ 2688 struct mutex compat_devs_mutex; 2689 /* Maintains compat devices for each net namespace */ 2690 struct xarray compat_devs; 2691 2692 /* Used by iWarp CM */ 2693 char iw_ifname[IFNAMSIZ]; 2694 u32 iw_driver_flags; 2695 }; 2696 2697 struct ib_client { 2698 const char *name; 2699 void (*add) (struct ib_device *); 2700 void (*remove)(struct ib_device *, void *client_data); 2701 void (*rename)(struct ib_device *dev, void *client_data); 2702 2703 /* Returns the net_dev belonging to this ib_client and matching the 2704 * given parameters. 2705 * @dev: An RDMA device that the net_dev use for communication. 2706 * @port: A physical port number on the RDMA device. 2707 * @pkey: P_Key that the net_dev uses if applicable. 2708 * @gid: A GID that the net_dev uses to communicate. 2709 * @addr: An IP address the net_dev is configured with. 2710 * @client_data: The device's client data set by ib_set_client_data(). 2711 * 2712 * An ib_client that implements a net_dev on top of RDMA devices 2713 * (such as IP over IB) should implement this callback, allowing the 2714 * rdma_cm module to find the right net_dev for a given request. 2715 * 2716 * The caller is responsible for calling dev_put on the returned 2717 * netdev. */ 2718 struct net_device *(*get_net_dev_by_params)( 2719 struct ib_device *dev, 2720 u8 port, 2721 u16 pkey, 2722 const union ib_gid *gid, 2723 const struct sockaddr *addr, 2724 void *client_data); 2725 struct list_head list; 2726 u32 client_id; 2727 2728 /* kverbs are not required by the client */ 2729 u8 no_kverbs_req:1; 2730 }; 2731 2732 /* 2733 * IB block DMA iterator 2734 * 2735 * Iterates the DMA-mapped SGL in contiguous memory blocks aligned 2736 * to a HW supported page size. 2737 */ 2738 struct ib_block_iter { 2739 /* internal states */ 2740 struct scatterlist *__sg; /* sg holding the current aligned block */ 2741 dma_addr_t __dma_addr; /* unaligned DMA address of this block */ 2742 unsigned int __sg_nents; /* number of SG entries */ 2743 unsigned int __sg_advance; /* number of bytes to advance in sg in next step */ 2744 unsigned int __pg_bit; /* alignment of current block */ 2745 }; 2746 2747 struct ib_device *_ib_alloc_device(size_t size); 2748 #define ib_alloc_device(drv_struct, member) \ 2749 container_of(_ib_alloc_device(sizeof(struct drv_struct) + \ 2750 BUILD_BUG_ON_ZERO(offsetof( \ 2751 struct drv_struct, member))), \ 2752 struct drv_struct, member) 2753 2754 void ib_dealloc_device(struct ib_device *device); 2755 2756 void ib_get_device_fw_str(struct ib_device *device, char *str); 2757 2758 int ib_register_device(struct ib_device *device, const char *name); 2759 void ib_unregister_device(struct ib_device *device); 2760 void ib_unregister_driver(enum rdma_driver_id driver_id); 2761 void ib_unregister_device_and_put(struct ib_device *device); 2762 void ib_unregister_device_queued(struct ib_device *ib_dev); 2763 2764 int ib_register_client (struct ib_client *client); 2765 void ib_unregister_client(struct ib_client *client); 2766 2767 void __rdma_block_iter_start(struct ib_block_iter *biter, 2768 struct scatterlist *sglist, 2769 unsigned int nents, 2770 unsigned long pgsz); 2771 bool __rdma_block_iter_next(struct ib_block_iter *biter); 2772 2773 /** 2774 * rdma_block_iter_dma_address - get the aligned dma address of the current 2775 * block held by the block iterator. 2776 * @biter: block iterator holding the memory block 2777 */ 2778 static inline dma_addr_t 2779 rdma_block_iter_dma_address(struct ib_block_iter *biter) 2780 { 2781 return biter->__dma_addr & ~(BIT_ULL(biter->__pg_bit) - 1); 2782 } 2783 2784 /** 2785 * rdma_for_each_block - iterate over contiguous memory blocks of the sg list 2786 * @sglist: sglist to iterate over 2787 * @biter: block iterator holding the memory block 2788 * @nents: maximum number of sg entries to iterate over 2789 * @pgsz: best HW supported page size to use 2790 * 2791 * Callers may use rdma_block_iter_dma_address() to get each 2792 * blocks aligned DMA address. 2793 */ 2794 #define rdma_for_each_block(sglist, biter, nents, pgsz) \ 2795 for (__rdma_block_iter_start(biter, sglist, nents, \ 2796 pgsz); \ 2797 __rdma_block_iter_next(biter);) 2798 2799 /** 2800 * ib_get_client_data - Get IB client context 2801 * @device:Device to get context for 2802 * @client:Client to get context for 2803 * 2804 * ib_get_client_data() returns the client context data set with 2805 * ib_set_client_data(). This can only be called while the client is 2806 * registered to the device, once the ib_client remove() callback returns this 2807 * cannot be called. 2808 */ 2809 static inline void *ib_get_client_data(struct ib_device *device, 2810 struct ib_client *client) 2811 { 2812 return xa_load(&device->client_data, client->client_id); 2813 } 2814 void ib_set_client_data(struct ib_device *device, struct ib_client *client, 2815 void *data); 2816 void ib_set_device_ops(struct ib_device *device, 2817 const struct ib_device_ops *ops); 2818 2819 #if IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS) 2820 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma, 2821 unsigned long pfn, unsigned long size, pgprot_t prot); 2822 #else 2823 static inline int rdma_user_mmap_io(struct ib_ucontext *ucontext, 2824 struct vm_area_struct *vma, 2825 unsigned long pfn, unsigned long size, 2826 pgprot_t prot) 2827 { 2828 return -EINVAL; 2829 } 2830 #endif 2831 2832 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len) 2833 { 2834 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0; 2835 } 2836 2837 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len) 2838 { 2839 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0; 2840 } 2841 2842 static inline bool ib_is_buffer_cleared(const void __user *p, 2843 size_t len) 2844 { 2845 bool ret; 2846 u8 *buf; 2847 2848 if (len > USHRT_MAX) 2849 return false; 2850 2851 buf = memdup_user(p, len); 2852 if (IS_ERR(buf)) 2853 return false; 2854 2855 ret = !memchr_inv(buf, 0, len); 2856 kfree(buf); 2857 return ret; 2858 } 2859 2860 static inline bool ib_is_udata_cleared(struct ib_udata *udata, 2861 size_t offset, 2862 size_t len) 2863 { 2864 return ib_is_buffer_cleared(udata->inbuf + offset, len); 2865 } 2866 2867 /** 2868 * ib_is_destroy_retryable - Check whether the uobject destruction 2869 * is retryable. 2870 * @ret: The initial destruction return code 2871 * @why: remove reason 2872 * @uobj: The uobject that is destroyed 2873 * 2874 * This function is a helper function that IB layer and low-level drivers 2875 * can use to consider whether the destruction of the given uobject is 2876 * retry-able. 2877 * It checks the original return code, if it wasn't success the destruction 2878 * is retryable according to the ucontext state (i.e. cleanup_retryable) and 2879 * the remove reason. (i.e. why). 2880 * Must be called with the object locked for destroy. 2881 */ 2882 static inline bool ib_is_destroy_retryable(int ret, enum rdma_remove_reason why, 2883 struct ib_uobject *uobj) 2884 { 2885 return ret && (why == RDMA_REMOVE_DESTROY || 2886 uobj->context->cleanup_retryable); 2887 } 2888 2889 /** 2890 * ib_destroy_usecnt - Called during destruction to check the usecnt 2891 * @usecnt: The usecnt atomic 2892 * @why: remove reason 2893 * @uobj: The uobject that is destroyed 2894 * 2895 * Non-zero usecnts will block destruction unless destruction was triggered by 2896 * a ucontext cleanup. 2897 */ 2898 static inline int ib_destroy_usecnt(atomic_t *usecnt, 2899 enum rdma_remove_reason why, 2900 struct ib_uobject *uobj) 2901 { 2902 if (atomic_read(usecnt) && ib_is_destroy_retryable(-EBUSY, why, uobj)) 2903 return -EBUSY; 2904 return 0; 2905 } 2906 2907 /** 2908 * ib_modify_qp_is_ok - Check that the supplied attribute mask 2909 * contains all required attributes and no attributes not allowed for 2910 * the given QP state transition. 2911 * @cur_state: Current QP state 2912 * @next_state: Next QP state 2913 * @type: QP type 2914 * @mask: Mask of supplied QP attributes 2915 * 2916 * This function is a helper function that a low-level driver's 2917 * modify_qp method can use to validate the consumer's input. It 2918 * checks that cur_state and next_state are valid QP states, that a 2919 * transition from cur_state to next_state is allowed by the IB spec, 2920 * and that the attribute mask supplied is allowed for the transition. 2921 */ 2922 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state, 2923 enum ib_qp_type type, enum ib_qp_attr_mask mask); 2924 2925 void ib_register_event_handler(struct ib_event_handler *event_handler); 2926 void ib_unregister_event_handler(struct ib_event_handler *event_handler); 2927 void ib_dispatch_event(struct ib_event *event); 2928 2929 int ib_query_port(struct ib_device *device, 2930 u8 port_num, struct ib_port_attr *port_attr); 2931 2932 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device, 2933 u8 port_num); 2934 2935 /** 2936 * rdma_cap_ib_switch - Check if the device is IB switch 2937 * @device: Device to check 2938 * 2939 * Device driver is responsible for setting is_switch bit on 2940 * in ib_device structure at init time. 2941 * 2942 * Return: true if the device is IB switch. 2943 */ 2944 static inline bool rdma_cap_ib_switch(const struct ib_device *device) 2945 { 2946 return device->is_switch; 2947 } 2948 2949 /** 2950 * rdma_start_port - Return the first valid port number for the device 2951 * specified 2952 * 2953 * @device: Device to be checked 2954 * 2955 * Return start port number 2956 */ 2957 static inline u8 rdma_start_port(const struct ib_device *device) 2958 { 2959 return rdma_cap_ib_switch(device) ? 0 : 1; 2960 } 2961 2962 /** 2963 * rdma_for_each_port - Iterate over all valid port numbers of the IB device 2964 * @device - The struct ib_device * to iterate over 2965 * @iter - The unsigned int to store the port number 2966 */ 2967 #define rdma_for_each_port(device, iter) \ 2968 for (iter = rdma_start_port(device + BUILD_BUG_ON_ZERO(!__same_type( \ 2969 unsigned int, iter))); \ 2970 iter <= rdma_end_port(device); (iter)++) 2971 2972 /** 2973 * rdma_end_port - Return the last valid port number for the device 2974 * specified 2975 * 2976 * @device: Device to be checked 2977 * 2978 * Return last port number 2979 */ 2980 static inline u8 rdma_end_port(const struct ib_device *device) 2981 { 2982 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt; 2983 } 2984 2985 static inline int rdma_is_port_valid(const struct ib_device *device, 2986 unsigned int port) 2987 { 2988 return (port >= rdma_start_port(device) && 2989 port <= rdma_end_port(device)); 2990 } 2991 2992 static inline bool rdma_is_grh_required(const struct ib_device *device, 2993 u8 port_num) 2994 { 2995 return device->port_data[port_num].immutable.core_cap_flags & 2996 RDMA_CORE_PORT_IB_GRH_REQUIRED; 2997 } 2998 2999 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num) 3000 { 3001 return device->port_data[port_num].immutable.core_cap_flags & 3002 RDMA_CORE_CAP_PROT_IB; 3003 } 3004 3005 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num) 3006 { 3007 return device->port_data[port_num].immutable.core_cap_flags & 3008 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP); 3009 } 3010 3011 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num) 3012 { 3013 return device->port_data[port_num].immutable.core_cap_flags & 3014 RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP; 3015 } 3016 3017 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num) 3018 { 3019 return device->port_data[port_num].immutable.core_cap_flags & 3020 RDMA_CORE_CAP_PROT_ROCE; 3021 } 3022 3023 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num) 3024 { 3025 return device->port_data[port_num].immutable.core_cap_flags & 3026 RDMA_CORE_CAP_PROT_IWARP; 3027 } 3028 3029 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num) 3030 { 3031 return rdma_protocol_ib(device, port_num) || 3032 rdma_protocol_roce(device, port_num); 3033 } 3034 3035 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num) 3036 { 3037 return device->port_data[port_num].immutable.core_cap_flags & 3038 RDMA_CORE_CAP_PROT_RAW_PACKET; 3039 } 3040 3041 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num) 3042 { 3043 return device->port_data[port_num].immutable.core_cap_flags & 3044 RDMA_CORE_CAP_PROT_USNIC; 3045 } 3046 3047 /** 3048 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband 3049 * Management Datagrams. 3050 * @device: Device to check 3051 * @port_num: Port number to check 3052 * 3053 * Management Datagrams (MAD) are a required part of the InfiniBand 3054 * specification and are supported on all InfiniBand devices. A slightly 3055 * extended version are also supported on OPA interfaces. 3056 * 3057 * Return: true if the port supports sending/receiving of MAD packets. 3058 */ 3059 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num) 3060 { 3061 return device->port_data[port_num].immutable.core_cap_flags & 3062 RDMA_CORE_CAP_IB_MAD; 3063 } 3064 3065 /** 3066 * rdma_cap_opa_mad - Check if the port of device provides support for OPA 3067 * Management Datagrams. 3068 * @device: Device to check 3069 * @port_num: Port number to check 3070 * 3071 * Intel OmniPath devices extend and/or replace the InfiniBand Management 3072 * datagrams with their own versions. These OPA MADs share many but not all of 3073 * the characteristics of InfiniBand MADs. 3074 * 3075 * OPA MADs differ in the following ways: 3076 * 3077 * 1) MADs are variable size up to 2K 3078 * IBTA defined MADs remain fixed at 256 bytes 3079 * 2) OPA SMPs must carry valid PKeys 3080 * 3) OPA SMP packets are a different format 3081 * 3082 * Return: true if the port supports OPA MAD packet formats. 3083 */ 3084 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num) 3085 { 3086 return device->port_data[port_num].immutable.core_cap_flags & 3087 RDMA_CORE_CAP_OPA_MAD; 3088 } 3089 3090 /** 3091 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband 3092 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI). 3093 * @device: Device to check 3094 * @port_num: Port number to check 3095 * 3096 * Each InfiniBand node is required to provide a Subnet Management Agent 3097 * that the subnet manager can access. Prior to the fabric being fully 3098 * configured by the subnet manager, the SMA is accessed via a well known 3099 * interface called the Subnet Management Interface (SMI). This interface 3100 * uses directed route packets to communicate with the SM to get around the 3101 * chicken and egg problem of the SM needing to know what's on the fabric 3102 * in order to configure the fabric, and needing to configure the fabric in 3103 * order to send packets to the devices on the fabric. These directed 3104 * route packets do not need the fabric fully configured in order to reach 3105 * their destination. The SMI is the only method allowed to send 3106 * directed route packets on an InfiniBand fabric. 3107 * 3108 * Return: true if the port provides an SMI. 3109 */ 3110 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num) 3111 { 3112 return device->port_data[port_num].immutable.core_cap_flags & 3113 RDMA_CORE_CAP_IB_SMI; 3114 } 3115 3116 /** 3117 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband 3118 * Communication Manager. 3119 * @device: Device to check 3120 * @port_num: Port number to check 3121 * 3122 * The InfiniBand Communication Manager is one of many pre-defined General 3123 * Service Agents (GSA) that are accessed via the General Service 3124 * Interface (GSI). It's role is to facilitate establishment of connections 3125 * between nodes as well as other management related tasks for established 3126 * connections. 3127 * 3128 * Return: true if the port supports an IB CM (this does not guarantee that 3129 * a CM is actually running however). 3130 */ 3131 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num) 3132 { 3133 return device->port_data[port_num].immutable.core_cap_flags & 3134 RDMA_CORE_CAP_IB_CM; 3135 } 3136 3137 /** 3138 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP 3139 * Communication Manager. 3140 * @device: Device to check 3141 * @port_num: Port number to check 3142 * 3143 * Similar to above, but specific to iWARP connections which have a different 3144 * managment protocol than InfiniBand. 3145 * 3146 * Return: true if the port supports an iWARP CM (this does not guarantee that 3147 * a CM is actually running however). 3148 */ 3149 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num) 3150 { 3151 return device->port_data[port_num].immutable.core_cap_flags & 3152 RDMA_CORE_CAP_IW_CM; 3153 } 3154 3155 /** 3156 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband 3157 * Subnet Administration. 3158 * @device: Device to check 3159 * @port_num: Port number to check 3160 * 3161 * An InfiniBand Subnet Administration (SA) service is a pre-defined General 3162 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand 3163 * fabrics, devices should resolve routes to other hosts by contacting the 3164 * SA to query the proper route. 3165 * 3166 * Return: true if the port should act as a client to the fabric Subnet 3167 * Administration interface. This does not imply that the SA service is 3168 * running locally. 3169 */ 3170 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num) 3171 { 3172 return device->port_data[port_num].immutable.core_cap_flags & 3173 RDMA_CORE_CAP_IB_SA; 3174 } 3175 3176 /** 3177 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband 3178 * Multicast. 3179 * @device: Device to check 3180 * @port_num: Port number to check 3181 * 3182 * InfiniBand multicast registration is more complex than normal IPv4 or 3183 * IPv6 multicast registration. Each Host Channel Adapter must register 3184 * with the Subnet Manager when it wishes to join a multicast group. It 3185 * should do so only once regardless of how many queue pairs it subscribes 3186 * to this group. And it should leave the group only after all queue pairs 3187 * attached to the group have been detached. 3188 * 3189 * Return: true if the port must undertake the additional adminstrative 3190 * overhead of registering/unregistering with the SM and tracking of the 3191 * total number of queue pairs attached to the multicast group. 3192 */ 3193 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num) 3194 { 3195 return rdma_cap_ib_sa(device, port_num); 3196 } 3197 3198 /** 3199 * rdma_cap_af_ib - Check if the port of device has the capability 3200 * Native Infiniband Address. 3201 * @device: Device to check 3202 * @port_num: Port number to check 3203 * 3204 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default 3205 * GID. RoCE uses a different mechanism, but still generates a GID via 3206 * a prescribed mechanism and port specific data. 3207 * 3208 * Return: true if the port uses a GID address to identify devices on the 3209 * network. 3210 */ 3211 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num) 3212 { 3213 return device->port_data[port_num].immutable.core_cap_flags & 3214 RDMA_CORE_CAP_AF_IB; 3215 } 3216 3217 /** 3218 * rdma_cap_eth_ah - Check if the port of device has the capability 3219 * Ethernet Address Handle. 3220 * @device: Device to check 3221 * @port_num: Port number to check 3222 * 3223 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique 3224 * to fabricate GIDs over Ethernet/IP specific addresses native to the 3225 * port. Normally, packet headers are generated by the sending host 3226 * adapter, but when sending connectionless datagrams, we must manually 3227 * inject the proper headers for the fabric we are communicating over. 3228 * 3229 * Return: true if we are running as a RoCE port and must force the 3230 * addition of a Global Route Header built from our Ethernet Address 3231 * Handle into our header list for connectionless packets. 3232 */ 3233 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num) 3234 { 3235 return device->port_data[port_num].immutable.core_cap_flags & 3236 RDMA_CORE_CAP_ETH_AH; 3237 } 3238 3239 /** 3240 * rdma_cap_opa_ah - Check if the port of device supports 3241 * OPA Address handles 3242 * @device: Device to check 3243 * @port_num: Port number to check 3244 * 3245 * Return: true if we are running on an OPA device which supports 3246 * the extended OPA addressing. 3247 */ 3248 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num) 3249 { 3250 return (device->port_data[port_num].immutable.core_cap_flags & 3251 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH; 3252 } 3253 3254 /** 3255 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port. 3256 * 3257 * @device: Device 3258 * @port_num: Port number 3259 * 3260 * This MAD size includes the MAD headers and MAD payload. No other headers 3261 * are included. 3262 * 3263 * Return the max MAD size required by the Port. Will return 0 if the port 3264 * does not support MADs 3265 */ 3266 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num) 3267 { 3268 return device->port_data[port_num].immutable.max_mad_size; 3269 } 3270 3271 /** 3272 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table 3273 * @device: Device to check 3274 * @port_num: Port number to check 3275 * 3276 * RoCE GID table mechanism manages the various GIDs for a device. 3277 * 3278 * NOTE: if allocating the port's GID table has failed, this call will still 3279 * return true, but any RoCE GID table API will fail. 3280 * 3281 * Return: true if the port uses RoCE GID table mechanism in order to manage 3282 * its GIDs. 3283 */ 3284 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device, 3285 u8 port_num) 3286 { 3287 return rdma_protocol_roce(device, port_num) && 3288 device->ops.add_gid && device->ops.del_gid; 3289 } 3290 3291 /* 3292 * Check if the device supports READ W/ INVALIDATE. 3293 */ 3294 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num) 3295 { 3296 /* 3297 * iWarp drivers must support READ W/ INVALIDATE. No other protocol 3298 * has support for it yet. 3299 */ 3300 return rdma_protocol_iwarp(dev, port_num); 3301 } 3302 3303 /** 3304 * rdma_find_pg_bit - Find page bit given address and HW supported page sizes 3305 * 3306 * @addr: address 3307 * @pgsz_bitmap: bitmap of HW supported page sizes 3308 */ 3309 static inline unsigned int rdma_find_pg_bit(unsigned long addr, 3310 unsigned long pgsz_bitmap) 3311 { 3312 unsigned long align; 3313 unsigned long pgsz; 3314 3315 align = addr & -addr; 3316 3317 /* Find page bit such that addr is aligned to the highest supported 3318 * HW page size 3319 */ 3320 pgsz = pgsz_bitmap & ~(-align << 1); 3321 if (!pgsz) 3322 return __ffs(pgsz_bitmap); 3323 3324 return __fls(pgsz); 3325 } 3326 3327 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port, 3328 int state); 3329 int ib_get_vf_config(struct ib_device *device, int vf, u8 port, 3330 struct ifla_vf_info *info); 3331 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port, 3332 struct ifla_vf_stats *stats); 3333 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid, 3334 int type); 3335 3336 int ib_query_pkey(struct ib_device *device, 3337 u8 port_num, u16 index, u16 *pkey); 3338 3339 int ib_modify_device(struct ib_device *device, 3340 int device_modify_mask, 3341 struct ib_device_modify *device_modify); 3342 3343 int ib_modify_port(struct ib_device *device, 3344 u8 port_num, int port_modify_mask, 3345 struct ib_port_modify *port_modify); 3346 3347 int ib_find_gid(struct ib_device *device, union ib_gid *gid, 3348 u8 *port_num, u16 *index); 3349 3350 int ib_find_pkey(struct ib_device *device, 3351 u8 port_num, u16 pkey, u16 *index); 3352 3353 enum ib_pd_flags { 3354 /* 3355 * Create a memory registration for all memory in the system and place 3356 * the rkey for it into pd->unsafe_global_rkey. This can be used by 3357 * ULPs to avoid the overhead of dynamic MRs. 3358 * 3359 * This flag is generally considered unsafe and must only be used in 3360 * extremly trusted environments. Every use of it will log a warning 3361 * in the kernel log. 3362 */ 3363 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01, 3364 }; 3365 3366 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags, 3367 const char *caller); 3368 3369 #define ib_alloc_pd(device, flags) \ 3370 __ib_alloc_pd((device), (flags), KBUILD_MODNAME) 3371 3372 /** 3373 * ib_dealloc_pd_user - Deallocate kernel/user PD 3374 * @pd: The protection domain 3375 * @udata: Valid user data or NULL for kernel objects 3376 */ 3377 void ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata); 3378 3379 /** 3380 * ib_dealloc_pd - Deallocate kernel PD 3381 * @pd: The protection domain 3382 * 3383 * NOTE: for user PD use ib_dealloc_pd_user with valid udata! 3384 */ 3385 static inline void ib_dealloc_pd(struct ib_pd *pd) 3386 { 3387 ib_dealloc_pd_user(pd, NULL); 3388 } 3389 3390 enum rdma_create_ah_flags { 3391 /* In a sleepable context */ 3392 RDMA_CREATE_AH_SLEEPABLE = BIT(0), 3393 }; 3394 3395 /** 3396 * rdma_create_ah - Creates an address handle for the given address vector. 3397 * @pd: The protection domain associated with the address handle. 3398 * @ah_attr: The attributes of the address vector. 3399 * @flags: Create address handle flags (see enum rdma_create_ah_flags). 3400 * 3401 * The address handle is used to reference a local or global destination 3402 * in all UD QP post sends. 3403 */ 3404 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr, 3405 u32 flags); 3406 3407 /** 3408 * rdma_create_user_ah - Creates an address handle for the given address vector. 3409 * It resolves destination mac address for ah attribute of RoCE type. 3410 * @pd: The protection domain associated with the address handle. 3411 * @ah_attr: The attributes of the address vector. 3412 * @udata: pointer to user's input output buffer information need by 3413 * provider driver. 3414 * 3415 * It returns 0 on success and returns appropriate error code on error. 3416 * The address handle is used to reference a local or global destination 3417 * in all UD QP post sends. 3418 */ 3419 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd, 3420 struct rdma_ah_attr *ah_attr, 3421 struct ib_udata *udata); 3422 /** 3423 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header 3424 * work completion. 3425 * @hdr: the L3 header to parse 3426 * @net_type: type of header to parse 3427 * @sgid: place to store source gid 3428 * @dgid: place to store destination gid 3429 */ 3430 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr, 3431 enum rdma_network_type net_type, 3432 union ib_gid *sgid, union ib_gid *dgid); 3433 3434 /** 3435 * ib_get_rdma_header_version - Get the header version 3436 * @hdr: the L3 header to parse 3437 */ 3438 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr); 3439 3440 /** 3441 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a 3442 * work completion. 3443 * @device: Device on which the received message arrived. 3444 * @port_num: Port on which the received message arrived. 3445 * @wc: Work completion associated with the received message. 3446 * @grh: References the received global route header. This parameter is 3447 * ignored unless the work completion indicates that the GRH is valid. 3448 * @ah_attr: Returned attributes that can be used when creating an address 3449 * handle for replying to the message. 3450 * When ib_init_ah_attr_from_wc() returns success, 3451 * (a) for IB link layer it optionally contains a reference to SGID attribute 3452 * when GRH is present for IB link layer. 3453 * (b) for RoCE link layer it contains a reference to SGID attribute. 3454 * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID 3455 * attributes which are initialized using ib_init_ah_attr_from_wc(). 3456 * 3457 */ 3458 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num, 3459 const struct ib_wc *wc, const struct ib_grh *grh, 3460 struct rdma_ah_attr *ah_attr); 3461 3462 /** 3463 * ib_create_ah_from_wc - Creates an address handle associated with the 3464 * sender of the specified work completion. 3465 * @pd: The protection domain associated with the address handle. 3466 * @wc: Work completion information associated with a received message. 3467 * @grh: References the received global route header. This parameter is 3468 * ignored unless the work completion indicates that the GRH is valid. 3469 * @port_num: The outbound port number to associate with the address. 3470 * 3471 * The address handle is used to reference a local or global destination 3472 * in all UD QP post sends. 3473 */ 3474 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc, 3475 const struct ib_grh *grh, u8 port_num); 3476 3477 /** 3478 * rdma_modify_ah - Modifies the address vector associated with an address 3479 * handle. 3480 * @ah: The address handle to modify. 3481 * @ah_attr: The new address vector attributes to associate with the 3482 * address handle. 3483 */ 3484 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr); 3485 3486 /** 3487 * rdma_query_ah - Queries the address vector associated with an address 3488 * handle. 3489 * @ah: The address handle to query. 3490 * @ah_attr: The address vector attributes associated with the address 3491 * handle. 3492 */ 3493 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr); 3494 3495 enum rdma_destroy_ah_flags { 3496 /* In a sleepable context */ 3497 RDMA_DESTROY_AH_SLEEPABLE = BIT(0), 3498 }; 3499 3500 /** 3501 * rdma_destroy_ah_user - Destroys an address handle. 3502 * @ah: The address handle to destroy. 3503 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags). 3504 * @udata: Valid user data or NULL for kernel objects 3505 */ 3506 int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata); 3507 3508 /** 3509 * rdma_destroy_ah - Destroys an kernel address handle. 3510 * @ah: The address handle to destroy. 3511 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags). 3512 * 3513 * NOTE: for user ah use rdma_destroy_ah_user with valid udata! 3514 */ 3515 static inline int rdma_destroy_ah(struct ib_ah *ah, u32 flags) 3516 { 3517 return rdma_destroy_ah_user(ah, flags, NULL); 3518 } 3519 3520 /** 3521 * ib_create_srq - Creates a SRQ associated with the specified protection 3522 * domain. 3523 * @pd: The protection domain associated with the SRQ. 3524 * @srq_init_attr: A list of initial attributes required to create the 3525 * SRQ. If SRQ creation succeeds, then the attributes are updated to 3526 * the actual capabilities of the created SRQ. 3527 * 3528 * srq_attr->max_wr and srq_attr->max_sge are read the determine the 3529 * requested size of the SRQ, and set to the actual values allocated 3530 * on return. If ib_create_srq() succeeds, then max_wr and max_sge 3531 * will always be at least as large as the requested values. 3532 */ 3533 struct ib_srq *ib_create_srq(struct ib_pd *pd, 3534 struct ib_srq_init_attr *srq_init_attr); 3535 3536 /** 3537 * ib_modify_srq - Modifies the attributes for the specified SRQ. 3538 * @srq: The SRQ to modify. 3539 * @srq_attr: On input, specifies the SRQ attributes to modify. On output, 3540 * the current values of selected SRQ attributes are returned. 3541 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ 3542 * are being modified. 3543 * 3544 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or 3545 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when 3546 * the number of receives queued drops below the limit. 3547 */ 3548 int ib_modify_srq(struct ib_srq *srq, 3549 struct ib_srq_attr *srq_attr, 3550 enum ib_srq_attr_mask srq_attr_mask); 3551 3552 /** 3553 * ib_query_srq - Returns the attribute list and current values for the 3554 * specified SRQ. 3555 * @srq: The SRQ to query. 3556 * @srq_attr: The attributes of the specified SRQ. 3557 */ 3558 int ib_query_srq(struct ib_srq *srq, 3559 struct ib_srq_attr *srq_attr); 3560 3561 /** 3562 * ib_destroy_srq_user - Destroys the specified SRQ. 3563 * @srq: The SRQ to destroy. 3564 * @udata: Valid user data or NULL for kernel objects 3565 */ 3566 int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata); 3567 3568 /** 3569 * ib_destroy_srq - Destroys the specified kernel SRQ. 3570 * @srq: The SRQ to destroy. 3571 * 3572 * NOTE: for user srq use ib_destroy_srq_user with valid udata! 3573 */ 3574 static inline int ib_destroy_srq(struct ib_srq *srq) 3575 { 3576 return ib_destroy_srq_user(srq, NULL); 3577 } 3578 3579 /** 3580 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ. 3581 * @srq: The SRQ to post the work request on. 3582 * @recv_wr: A list of work requests to post on the receive queue. 3583 * @bad_recv_wr: On an immediate failure, this parameter will reference 3584 * the work request that failed to be posted on the QP. 3585 */ 3586 static inline int ib_post_srq_recv(struct ib_srq *srq, 3587 const struct ib_recv_wr *recv_wr, 3588 const struct ib_recv_wr **bad_recv_wr) 3589 { 3590 const struct ib_recv_wr *dummy; 3591 3592 return srq->device->ops.post_srq_recv(srq, recv_wr, 3593 bad_recv_wr ? : &dummy); 3594 } 3595 3596 /** 3597 * ib_create_qp_user - Creates a QP associated with the specified protection 3598 * domain. 3599 * @pd: The protection domain associated with the QP. 3600 * @qp_init_attr: A list of initial attributes required to create the 3601 * QP. If QP creation succeeds, then the attributes are updated to 3602 * the actual capabilities of the created QP. 3603 * @udata: Valid user data or NULL for kernel objects 3604 */ 3605 struct ib_qp *ib_create_qp_user(struct ib_pd *pd, 3606 struct ib_qp_init_attr *qp_init_attr, 3607 struct ib_udata *udata); 3608 3609 /** 3610 * ib_create_qp - Creates a kernel QP associated with the specified protection 3611 * domain. 3612 * @pd: The protection domain associated with the QP. 3613 * @qp_init_attr: A list of initial attributes required to create the 3614 * QP. If QP creation succeeds, then the attributes are updated to 3615 * the actual capabilities of the created QP. 3616 * @udata: Valid user data or NULL for kernel objects 3617 * 3618 * NOTE: for user qp use ib_create_qp_user with valid udata! 3619 */ 3620 static inline struct ib_qp *ib_create_qp(struct ib_pd *pd, 3621 struct ib_qp_init_attr *qp_init_attr) 3622 { 3623 return ib_create_qp_user(pd, qp_init_attr, NULL); 3624 } 3625 3626 /** 3627 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP. 3628 * @qp: The QP to modify. 3629 * @attr: On input, specifies the QP attributes to modify. On output, 3630 * the current values of selected QP attributes are returned. 3631 * @attr_mask: A bit-mask used to specify which attributes of the QP 3632 * are being modified. 3633 * @udata: pointer to user's input output buffer information 3634 * are being modified. 3635 * It returns 0 on success and returns appropriate error code on error. 3636 */ 3637 int ib_modify_qp_with_udata(struct ib_qp *qp, 3638 struct ib_qp_attr *attr, 3639 int attr_mask, 3640 struct ib_udata *udata); 3641 3642 /** 3643 * ib_modify_qp - Modifies the attributes for the specified QP and then 3644 * transitions the QP to the given state. 3645 * @qp: The QP to modify. 3646 * @qp_attr: On input, specifies the QP attributes to modify. On output, 3647 * the current values of selected QP attributes are returned. 3648 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP 3649 * are being modified. 3650 */ 3651 int ib_modify_qp(struct ib_qp *qp, 3652 struct ib_qp_attr *qp_attr, 3653 int qp_attr_mask); 3654 3655 /** 3656 * ib_query_qp - Returns the attribute list and current values for the 3657 * specified QP. 3658 * @qp: The QP to query. 3659 * @qp_attr: The attributes of the specified QP. 3660 * @qp_attr_mask: A bit-mask used to select specific attributes to query. 3661 * @qp_init_attr: Additional attributes of the selected QP. 3662 * 3663 * The qp_attr_mask may be used to limit the query to gathering only the 3664 * selected attributes. 3665 */ 3666 int ib_query_qp(struct ib_qp *qp, 3667 struct ib_qp_attr *qp_attr, 3668 int qp_attr_mask, 3669 struct ib_qp_init_attr *qp_init_attr); 3670 3671 /** 3672 * ib_destroy_qp - Destroys the specified QP. 3673 * @qp: The QP to destroy. 3674 * @udata: Valid udata or NULL for kernel objects 3675 */ 3676 int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata); 3677 3678 /** 3679 * ib_destroy_qp - Destroys the specified kernel QP. 3680 * @qp: The QP to destroy. 3681 * 3682 * NOTE: for user qp use ib_destroy_qp_user with valid udata! 3683 */ 3684 static inline int ib_destroy_qp(struct ib_qp *qp) 3685 { 3686 return ib_destroy_qp_user(qp, NULL); 3687 } 3688 3689 /** 3690 * ib_open_qp - Obtain a reference to an existing sharable QP. 3691 * @xrcd - XRC domain 3692 * @qp_open_attr: Attributes identifying the QP to open. 3693 * 3694 * Returns a reference to a sharable QP. 3695 */ 3696 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd, 3697 struct ib_qp_open_attr *qp_open_attr); 3698 3699 /** 3700 * ib_close_qp - Release an external reference to a QP. 3701 * @qp: The QP handle to release 3702 * 3703 * The opened QP handle is released by the caller. The underlying 3704 * shared QP is not destroyed until all internal references are released. 3705 */ 3706 int ib_close_qp(struct ib_qp *qp); 3707 3708 /** 3709 * ib_post_send - Posts a list of work requests to the send queue of 3710 * the specified QP. 3711 * @qp: The QP to post the work request on. 3712 * @send_wr: A list of work requests to post on the send queue. 3713 * @bad_send_wr: On an immediate failure, this parameter will reference 3714 * the work request that failed to be posted on the QP. 3715 * 3716 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate 3717 * error is returned, the QP state shall not be affected, 3718 * ib_post_send() will return an immediate error after queueing any 3719 * earlier work requests in the list. 3720 */ 3721 static inline int ib_post_send(struct ib_qp *qp, 3722 const struct ib_send_wr *send_wr, 3723 const struct ib_send_wr **bad_send_wr) 3724 { 3725 const struct ib_send_wr *dummy; 3726 3727 return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy); 3728 } 3729 3730 /** 3731 * ib_post_recv - Posts a list of work requests to the receive queue of 3732 * the specified QP. 3733 * @qp: The QP to post the work request on. 3734 * @recv_wr: A list of work requests to post on the receive queue. 3735 * @bad_recv_wr: On an immediate failure, this parameter will reference 3736 * the work request that failed to be posted on the QP. 3737 */ 3738 static inline int ib_post_recv(struct ib_qp *qp, 3739 const struct ib_recv_wr *recv_wr, 3740 const struct ib_recv_wr **bad_recv_wr) 3741 { 3742 const struct ib_recv_wr *dummy; 3743 3744 return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy); 3745 } 3746 3747 struct ib_cq *__ib_alloc_cq_user(struct ib_device *dev, void *private, 3748 int nr_cqe, int comp_vector, 3749 enum ib_poll_context poll_ctx, 3750 const char *caller, struct ib_udata *udata); 3751 3752 /** 3753 * ib_alloc_cq_user: Allocate kernel/user CQ 3754 * @dev: The IB device 3755 * @private: Private data attached to the CQE 3756 * @nr_cqe: Number of CQEs in the CQ 3757 * @comp_vector: Completion vector used for the IRQs 3758 * @poll_ctx: Context used for polling the CQ 3759 * @udata: Valid user data or NULL for kernel objects 3760 */ 3761 static inline struct ib_cq *ib_alloc_cq_user(struct ib_device *dev, 3762 void *private, int nr_cqe, 3763 int comp_vector, 3764 enum ib_poll_context poll_ctx, 3765 struct ib_udata *udata) 3766 { 3767 return __ib_alloc_cq_user(dev, private, nr_cqe, comp_vector, poll_ctx, 3768 KBUILD_MODNAME, udata); 3769 } 3770 3771 /** 3772 * ib_alloc_cq: Allocate kernel CQ 3773 * @dev: The IB device 3774 * @private: Private data attached to the CQE 3775 * @nr_cqe: Number of CQEs in the CQ 3776 * @comp_vector: Completion vector used for the IRQs 3777 * @poll_ctx: Context used for polling the CQ 3778 * 3779 * NOTE: for user cq use ib_alloc_cq_user with valid udata! 3780 */ 3781 static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private, 3782 int nr_cqe, int comp_vector, 3783 enum ib_poll_context poll_ctx) 3784 { 3785 return ib_alloc_cq_user(dev, private, nr_cqe, comp_vector, poll_ctx, 3786 NULL); 3787 } 3788 3789 /** 3790 * ib_free_cq_user - Free kernel/user CQ 3791 * @cq: The CQ to free 3792 * @udata: Valid user data or NULL for kernel objects 3793 */ 3794 void ib_free_cq_user(struct ib_cq *cq, struct ib_udata *udata); 3795 3796 /** 3797 * ib_free_cq - Free kernel CQ 3798 * @cq: The CQ to free 3799 * 3800 * NOTE: for user cq use ib_free_cq_user with valid udata! 3801 */ 3802 static inline void ib_free_cq(struct ib_cq *cq) 3803 { 3804 ib_free_cq_user(cq, NULL); 3805 } 3806 3807 int ib_process_cq_direct(struct ib_cq *cq, int budget); 3808 3809 /** 3810 * ib_create_cq - Creates a CQ on the specified device. 3811 * @device: The device on which to create the CQ. 3812 * @comp_handler: A user-specified callback that is invoked when a 3813 * completion event occurs on the CQ. 3814 * @event_handler: A user-specified callback that is invoked when an 3815 * asynchronous event not associated with a completion occurs on the CQ. 3816 * @cq_context: Context associated with the CQ returned to the user via 3817 * the associated completion and event handlers. 3818 * @cq_attr: The attributes the CQ should be created upon. 3819 * 3820 * Users can examine the cq structure to determine the actual CQ size. 3821 */ 3822 struct ib_cq *__ib_create_cq(struct ib_device *device, 3823 ib_comp_handler comp_handler, 3824 void (*event_handler)(struct ib_event *, void *), 3825 void *cq_context, 3826 const struct ib_cq_init_attr *cq_attr, 3827 const char *caller); 3828 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \ 3829 __ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME) 3830 3831 /** 3832 * ib_resize_cq - Modifies the capacity of the CQ. 3833 * @cq: The CQ to resize. 3834 * @cqe: The minimum size of the CQ. 3835 * 3836 * Users can examine the cq structure to determine the actual CQ size. 3837 */ 3838 int ib_resize_cq(struct ib_cq *cq, int cqe); 3839 3840 /** 3841 * rdma_set_cq_moderation - Modifies moderation params of the CQ 3842 * @cq: The CQ to modify. 3843 * @cq_count: number of CQEs that will trigger an event 3844 * @cq_period: max period of time in usec before triggering an event 3845 * 3846 */ 3847 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period); 3848 3849 /** 3850 * ib_destroy_cq_user - Destroys the specified CQ. 3851 * @cq: The CQ to destroy. 3852 * @udata: Valid user data or NULL for kernel objects 3853 */ 3854 int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata); 3855 3856 /** 3857 * ib_destroy_cq - Destroys the specified kernel CQ. 3858 * @cq: The CQ to destroy. 3859 * 3860 * NOTE: for user cq use ib_destroy_cq_user with valid udata! 3861 */ 3862 static inline int ib_destroy_cq(struct ib_cq *cq) 3863 { 3864 return ib_destroy_cq_user(cq, NULL); 3865 } 3866 3867 /** 3868 * ib_poll_cq - poll a CQ for completion(s) 3869 * @cq:the CQ being polled 3870 * @num_entries:maximum number of completions to return 3871 * @wc:array of at least @num_entries &struct ib_wc where completions 3872 * will be returned 3873 * 3874 * Poll a CQ for (possibly multiple) completions. If the return value 3875 * is < 0, an error occurred. If the return value is >= 0, it is the 3876 * number of completions returned. If the return value is 3877 * non-negative and < num_entries, then the CQ was emptied. 3878 */ 3879 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries, 3880 struct ib_wc *wc) 3881 { 3882 return cq->device->ops.poll_cq(cq, num_entries, wc); 3883 } 3884 3885 /** 3886 * ib_req_notify_cq - Request completion notification on a CQ. 3887 * @cq: The CQ to generate an event for. 3888 * @flags: 3889 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP 3890 * to request an event on the next solicited event or next work 3891 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS 3892 * may also be |ed in to request a hint about missed events, as 3893 * described below. 3894 * 3895 * Return Value: 3896 * < 0 means an error occurred while requesting notification 3897 * == 0 means notification was requested successfully, and if 3898 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events 3899 * were missed and it is safe to wait for another event. In 3900 * this case is it guaranteed that any work completions added 3901 * to the CQ since the last CQ poll will trigger a completion 3902 * notification event. 3903 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed 3904 * in. It means that the consumer must poll the CQ again to 3905 * make sure it is empty to avoid missing an event because of a 3906 * race between requesting notification and an entry being 3907 * added to the CQ. This return value means it is possible 3908 * (but not guaranteed) that a work completion has been added 3909 * to the CQ since the last poll without triggering a 3910 * completion notification event. 3911 */ 3912 static inline int ib_req_notify_cq(struct ib_cq *cq, 3913 enum ib_cq_notify_flags flags) 3914 { 3915 return cq->device->ops.req_notify_cq(cq, flags); 3916 } 3917 3918 /** 3919 * ib_req_ncomp_notif - Request completion notification when there are 3920 * at least the specified number of unreaped completions on the CQ. 3921 * @cq: The CQ to generate an event for. 3922 * @wc_cnt: The number of unreaped completions that should be on the 3923 * CQ before an event is generated. 3924 */ 3925 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt) 3926 { 3927 return cq->device->ops.req_ncomp_notif ? 3928 cq->device->ops.req_ncomp_notif(cq, wc_cnt) : 3929 -ENOSYS; 3930 } 3931 3932 /** 3933 * ib_dma_mapping_error - check a DMA addr for error 3934 * @dev: The device for which the dma_addr was created 3935 * @dma_addr: The DMA address to check 3936 */ 3937 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr) 3938 { 3939 return dma_mapping_error(dev->dma_device, dma_addr); 3940 } 3941 3942 /** 3943 * ib_dma_map_single - Map a kernel virtual address to DMA address 3944 * @dev: The device for which the dma_addr is to be created 3945 * @cpu_addr: The kernel virtual address 3946 * @size: The size of the region in bytes 3947 * @direction: The direction of the DMA 3948 */ 3949 static inline u64 ib_dma_map_single(struct ib_device *dev, 3950 void *cpu_addr, size_t size, 3951 enum dma_data_direction direction) 3952 { 3953 return dma_map_single(dev->dma_device, cpu_addr, size, direction); 3954 } 3955 3956 /** 3957 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single() 3958 * @dev: The device for which the DMA address was created 3959 * @addr: The DMA address 3960 * @size: The size of the region in bytes 3961 * @direction: The direction of the DMA 3962 */ 3963 static inline void ib_dma_unmap_single(struct ib_device *dev, 3964 u64 addr, size_t size, 3965 enum dma_data_direction direction) 3966 { 3967 dma_unmap_single(dev->dma_device, addr, size, direction); 3968 } 3969 3970 /** 3971 * ib_dma_map_page - Map a physical page to DMA address 3972 * @dev: The device for which the dma_addr is to be created 3973 * @page: The page to be mapped 3974 * @offset: The offset within the page 3975 * @size: The size of the region in bytes 3976 * @direction: The direction of the DMA 3977 */ 3978 static inline u64 ib_dma_map_page(struct ib_device *dev, 3979 struct page *page, 3980 unsigned long offset, 3981 size_t size, 3982 enum dma_data_direction direction) 3983 { 3984 return dma_map_page(dev->dma_device, page, offset, size, direction); 3985 } 3986 3987 /** 3988 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page() 3989 * @dev: The device for which the DMA address was created 3990 * @addr: The DMA address 3991 * @size: The size of the region in bytes 3992 * @direction: The direction of the DMA 3993 */ 3994 static inline void ib_dma_unmap_page(struct ib_device *dev, 3995 u64 addr, size_t size, 3996 enum dma_data_direction direction) 3997 { 3998 dma_unmap_page(dev->dma_device, addr, size, direction); 3999 } 4000 4001 /** 4002 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses 4003 * @dev: The device for which the DMA addresses are to be created 4004 * @sg: The array of scatter/gather entries 4005 * @nents: The number of scatter/gather entries 4006 * @direction: The direction of the DMA 4007 */ 4008 static inline int ib_dma_map_sg(struct ib_device *dev, 4009 struct scatterlist *sg, int nents, 4010 enum dma_data_direction direction) 4011 { 4012 return dma_map_sg(dev->dma_device, sg, nents, direction); 4013 } 4014 4015 /** 4016 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses 4017 * @dev: The device for which the DMA addresses were created 4018 * @sg: The array of scatter/gather entries 4019 * @nents: The number of scatter/gather entries 4020 * @direction: The direction of the DMA 4021 */ 4022 static inline void ib_dma_unmap_sg(struct ib_device *dev, 4023 struct scatterlist *sg, int nents, 4024 enum dma_data_direction direction) 4025 { 4026 dma_unmap_sg(dev->dma_device, sg, nents, direction); 4027 } 4028 4029 static inline int ib_dma_map_sg_attrs(struct ib_device *dev, 4030 struct scatterlist *sg, int nents, 4031 enum dma_data_direction direction, 4032 unsigned long dma_attrs) 4033 { 4034 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, 4035 dma_attrs); 4036 } 4037 4038 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev, 4039 struct scatterlist *sg, int nents, 4040 enum dma_data_direction direction, 4041 unsigned long dma_attrs) 4042 { 4043 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs); 4044 } 4045 4046 /** 4047 * ib_dma_max_seg_size - Return the size limit of a single DMA transfer 4048 * @dev: The device to query 4049 * 4050 * The returned value represents a size in bytes. 4051 */ 4052 static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev) 4053 { 4054 struct device_dma_parameters *p = dev->dma_device->dma_parms; 4055 4056 return p ? p->max_segment_size : UINT_MAX; 4057 } 4058 4059 /** 4060 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU 4061 * @dev: The device for which the DMA address was created 4062 * @addr: The DMA address 4063 * @size: The size of the region in bytes 4064 * @dir: The direction of the DMA 4065 */ 4066 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev, 4067 u64 addr, 4068 size_t size, 4069 enum dma_data_direction dir) 4070 { 4071 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir); 4072 } 4073 4074 /** 4075 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device 4076 * @dev: The device for which the DMA address was created 4077 * @addr: The DMA address 4078 * @size: The size of the region in bytes 4079 * @dir: The direction of the DMA 4080 */ 4081 static inline void ib_dma_sync_single_for_device(struct ib_device *dev, 4082 u64 addr, 4083 size_t size, 4084 enum dma_data_direction dir) 4085 { 4086 dma_sync_single_for_device(dev->dma_device, addr, size, dir); 4087 } 4088 4089 /** 4090 * ib_dma_alloc_coherent - Allocate memory and map it for DMA 4091 * @dev: The device for which the DMA address is requested 4092 * @size: The size of the region to allocate in bytes 4093 * @dma_handle: A pointer for returning the DMA address of the region 4094 * @flag: memory allocator flags 4095 */ 4096 static inline void *ib_dma_alloc_coherent(struct ib_device *dev, 4097 size_t size, 4098 dma_addr_t *dma_handle, 4099 gfp_t flag) 4100 { 4101 return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag); 4102 } 4103 4104 /** 4105 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent() 4106 * @dev: The device for which the DMA addresses were allocated 4107 * @size: The size of the region 4108 * @cpu_addr: the address returned by ib_dma_alloc_coherent() 4109 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent() 4110 */ 4111 static inline void ib_dma_free_coherent(struct ib_device *dev, 4112 size_t size, void *cpu_addr, 4113 dma_addr_t dma_handle) 4114 { 4115 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle); 4116 } 4117 4118 /** 4119 * ib_dereg_mr_user - Deregisters a memory region and removes it from the 4120 * HCA translation table. 4121 * @mr: The memory region to deregister. 4122 * @udata: Valid user data or NULL for kernel object 4123 * 4124 * This function can fail, if the memory region has memory windows bound to it. 4125 */ 4126 int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata); 4127 4128 /** 4129 * ib_dereg_mr - Deregisters a kernel memory region and removes it from the 4130 * HCA translation table. 4131 * @mr: The memory region to deregister. 4132 * 4133 * This function can fail, if the memory region has memory windows bound to it. 4134 * 4135 * NOTE: for user mr use ib_dereg_mr_user with valid udata! 4136 */ 4137 static inline int ib_dereg_mr(struct ib_mr *mr) 4138 { 4139 return ib_dereg_mr_user(mr, NULL); 4140 } 4141 4142 struct ib_mr *ib_alloc_mr_user(struct ib_pd *pd, enum ib_mr_type mr_type, 4143 u32 max_num_sg, struct ib_udata *udata); 4144 4145 static inline struct ib_mr *ib_alloc_mr(struct ib_pd *pd, 4146 enum ib_mr_type mr_type, u32 max_num_sg) 4147 { 4148 return ib_alloc_mr_user(pd, mr_type, max_num_sg, NULL); 4149 } 4150 4151 /** 4152 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR 4153 * R_Key and L_Key. 4154 * @mr - struct ib_mr pointer to be updated. 4155 * @newkey - new key to be used. 4156 */ 4157 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey) 4158 { 4159 mr->lkey = (mr->lkey & 0xffffff00) | newkey; 4160 mr->rkey = (mr->rkey & 0xffffff00) | newkey; 4161 } 4162 4163 /** 4164 * ib_inc_rkey - increments the key portion of the given rkey. Can be used 4165 * for calculating a new rkey for type 2 memory windows. 4166 * @rkey - the rkey to increment. 4167 */ 4168 static inline u32 ib_inc_rkey(u32 rkey) 4169 { 4170 const u32 mask = 0x000000ff; 4171 return ((rkey + 1) & mask) | (rkey & ~mask); 4172 } 4173 4174 /** 4175 * ib_alloc_fmr - Allocates a unmapped fast memory region. 4176 * @pd: The protection domain associated with the unmapped region. 4177 * @mr_access_flags: Specifies the memory access rights. 4178 * @fmr_attr: Attributes of the unmapped region. 4179 * 4180 * A fast memory region must be mapped before it can be used as part of 4181 * a work request. 4182 */ 4183 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd, 4184 int mr_access_flags, 4185 struct ib_fmr_attr *fmr_attr); 4186 4187 /** 4188 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region. 4189 * @fmr: The fast memory region to associate with the pages. 4190 * @page_list: An array of physical pages to map to the fast memory region. 4191 * @list_len: The number of pages in page_list. 4192 * @iova: The I/O virtual address to use with the mapped region. 4193 */ 4194 static inline int ib_map_phys_fmr(struct ib_fmr *fmr, 4195 u64 *page_list, int list_len, 4196 u64 iova) 4197 { 4198 return fmr->device->ops.map_phys_fmr(fmr, page_list, list_len, iova); 4199 } 4200 4201 /** 4202 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions. 4203 * @fmr_list: A linked list of fast memory regions to unmap. 4204 */ 4205 int ib_unmap_fmr(struct list_head *fmr_list); 4206 4207 /** 4208 * ib_dealloc_fmr - Deallocates a fast memory region. 4209 * @fmr: The fast memory region to deallocate. 4210 */ 4211 int ib_dealloc_fmr(struct ib_fmr *fmr); 4212 4213 /** 4214 * ib_attach_mcast - Attaches the specified QP to a multicast group. 4215 * @qp: QP to attach to the multicast group. The QP must be type 4216 * IB_QPT_UD. 4217 * @gid: Multicast group GID. 4218 * @lid: Multicast group LID in host byte order. 4219 * 4220 * In order to send and receive multicast packets, subnet 4221 * administration must have created the multicast group and configured 4222 * the fabric appropriately. The port associated with the specified 4223 * QP must also be a member of the multicast group. 4224 */ 4225 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid); 4226 4227 /** 4228 * ib_detach_mcast - Detaches the specified QP from a multicast group. 4229 * @qp: QP to detach from the multicast group. 4230 * @gid: Multicast group GID. 4231 * @lid: Multicast group LID in host byte order. 4232 */ 4233 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid); 4234 4235 /** 4236 * ib_alloc_xrcd - Allocates an XRC domain. 4237 * @device: The device on which to allocate the XRC domain. 4238 * @caller: Module name for kernel consumers 4239 */ 4240 struct ib_xrcd *__ib_alloc_xrcd(struct ib_device *device, const char *caller); 4241 #define ib_alloc_xrcd(device) \ 4242 __ib_alloc_xrcd((device), KBUILD_MODNAME) 4243 4244 /** 4245 * ib_dealloc_xrcd - Deallocates an XRC domain. 4246 * @xrcd: The XRC domain to deallocate. 4247 * @udata: Valid user data or NULL for kernel object 4248 */ 4249 int ib_dealloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata); 4250 4251 static inline int ib_check_mr_access(int flags) 4252 { 4253 /* 4254 * Local write permission is required if remote write or 4255 * remote atomic permission is also requested. 4256 */ 4257 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) && 4258 !(flags & IB_ACCESS_LOCAL_WRITE)) 4259 return -EINVAL; 4260 4261 return 0; 4262 } 4263 4264 static inline bool ib_access_writable(int access_flags) 4265 { 4266 /* 4267 * We have writable memory backing the MR if any of the following 4268 * access flags are set. "Local write" and "remote write" obviously 4269 * require write access. "Remote atomic" can do things like fetch and 4270 * add, which will modify memory, and "MW bind" can change permissions 4271 * by binding a window. 4272 */ 4273 return access_flags & 4274 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE | 4275 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND); 4276 } 4277 4278 /** 4279 * ib_check_mr_status: lightweight check of MR status. 4280 * This routine may provide status checks on a selected 4281 * ib_mr. first use is for signature status check. 4282 * 4283 * @mr: A memory region. 4284 * @check_mask: Bitmask of which checks to perform from 4285 * ib_mr_status_check enumeration. 4286 * @mr_status: The container of relevant status checks. 4287 * failed checks will be indicated in the status bitmask 4288 * and the relevant info shall be in the error item. 4289 */ 4290 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask, 4291 struct ib_mr_status *mr_status); 4292 4293 /** 4294 * ib_device_try_get: Hold a registration lock 4295 * device: The device to lock 4296 * 4297 * A device under an active registration lock cannot become unregistered. It 4298 * is only possible to obtain a registration lock on a device that is fully 4299 * registered, otherwise this function returns false. 4300 * 4301 * The registration lock is only necessary for actions which require the 4302 * device to still be registered. Uses that only require the device pointer to 4303 * be valid should use get_device(&ibdev->dev) to hold the memory. 4304 * 4305 */ 4306 static inline bool ib_device_try_get(struct ib_device *dev) 4307 { 4308 return refcount_inc_not_zero(&dev->refcount); 4309 } 4310 4311 void ib_device_put(struct ib_device *device); 4312 struct ib_device *ib_device_get_by_netdev(struct net_device *ndev, 4313 enum rdma_driver_id driver_id); 4314 struct ib_device *ib_device_get_by_name(const char *name, 4315 enum rdma_driver_id driver_id); 4316 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port, 4317 u16 pkey, const union ib_gid *gid, 4318 const struct sockaddr *addr); 4319 int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev, 4320 unsigned int port); 4321 struct net_device *ib_device_netdev(struct ib_device *dev, u8 port); 4322 4323 struct ib_wq *ib_create_wq(struct ib_pd *pd, 4324 struct ib_wq_init_attr *init_attr); 4325 int ib_destroy_wq(struct ib_wq *wq, struct ib_udata *udata); 4326 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr, 4327 u32 wq_attr_mask); 4328 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device, 4329 struct ib_rwq_ind_table_init_attr* 4330 wq_ind_table_init_attr); 4331 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table); 4332 4333 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents, 4334 unsigned int *sg_offset, unsigned int page_size); 4335 4336 static inline int 4337 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents, 4338 unsigned int *sg_offset, unsigned int page_size) 4339 { 4340 int n; 4341 4342 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size); 4343 mr->iova = 0; 4344 4345 return n; 4346 } 4347 4348 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents, 4349 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64)); 4350 4351 void ib_drain_rq(struct ib_qp *qp); 4352 void ib_drain_sq(struct ib_qp *qp); 4353 void ib_drain_qp(struct ib_qp *qp); 4354 4355 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width); 4356 4357 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr) 4358 { 4359 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE) 4360 return attr->roce.dmac; 4361 return NULL; 4362 } 4363 4364 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid) 4365 { 4366 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 4367 attr->ib.dlid = (u16)dlid; 4368 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 4369 attr->opa.dlid = dlid; 4370 } 4371 4372 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr) 4373 { 4374 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 4375 return attr->ib.dlid; 4376 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 4377 return attr->opa.dlid; 4378 return 0; 4379 } 4380 4381 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl) 4382 { 4383 attr->sl = sl; 4384 } 4385 4386 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr) 4387 { 4388 return attr->sl; 4389 } 4390 4391 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr, 4392 u8 src_path_bits) 4393 { 4394 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 4395 attr->ib.src_path_bits = src_path_bits; 4396 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 4397 attr->opa.src_path_bits = src_path_bits; 4398 } 4399 4400 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr) 4401 { 4402 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 4403 return attr->ib.src_path_bits; 4404 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 4405 return attr->opa.src_path_bits; 4406 return 0; 4407 } 4408 4409 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr, 4410 bool make_grd) 4411 { 4412 if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 4413 attr->opa.make_grd = make_grd; 4414 } 4415 4416 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr) 4417 { 4418 if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 4419 return attr->opa.make_grd; 4420 return false; 4421 } 4422 4423 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num) 4424 { 4425 attr->port_num = port_num; 4426 } 4427 4428 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr) 4429 { 4430 return attr->port_num; 4431 } 4432 4433 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr, 4434 u8 static_rate) 4435 { 4436 attr->static_rate = static_rate; 4437 } 4438 4439 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr) 4440 { 4441 return attr->static_rate; 4442 } 4443 4444 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr, 4445 enum ib_ah_flags flag) 4446 { 4447 attr->ah_flags = flag; 4448 } 4449 4450 static inline enum ib_ah_flags 4451 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr) 4452 { 4453 return attr->ah_flags; 4454 } 4455 4456 static inline const struct ib_global_route 4457 *rdma_ah_read_grh(const struct rdma_ah_attr *attr) 4458 { 4459 return &attr->grh; 4460 } 4461 4462 /*To retrieve and modify the grh */ 4463 static inline struct ib_global_route 4464 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr) 4465 { 4466 return &attr->grh; 4467 } 4468 4469 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid) 4470 { 4471 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 4472 4473 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid)); 4474 } 4475 4476 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr, 4477 __be64 prefix) 4478 { 4479 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 4480 4481 grh->dgid.global.subnet_prefix = prefix; 4482 } 4483 4484 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr, 4485 __be64 if_id) 4486 { 4487 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 4488 4489 grh->dgid.global.interface_id = if_id; 4490 } 4491 4492 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr, 4493 union ib_gid *dgid, u32 flow_label, 4494 u8 sgid_index, u8 hop_limit, 4495 u8 traffic_class) 4496 { 4497 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 4498 4499 attr->ah_flags = IB_AH_GRH; 4500 if (dgid) 4501 grh->dgid = *dgid; 4502 grh->flow_label = flow_label; 4503 grh->sgid_index = sgid_index; 4504 grh->hop_limit = hop_limit; 4505 grh->traffic_class = traffic_class; 4506 grh->sgid_attr = NULL; 4507 } 4508 4509 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr); 4510 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid, 4511 u32 flow_label, u8 hop_limit, u8 traffic_class, 4512 const struct ib_gid_attr *sgid_attr); 4513 void rdma_copy_ah_attr(struct rdma_ah_attr *dest, 4514 const struct rdma_ah_attr *src); 4515 void rdma_replace_ah_attr(struct rdma_ah_attr *old, 4516 const struct rdma_ah_attr *new); 4517 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src); 4518 4519 /** 4520 * rdma_ah_find_type - Return address handle type. 4521 * 4522 * @dev: Device to be checked 4523 * @port_num: Port number 4524 */ 4525 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev, 4526 u8 port_num) 4527 { 4528 if (rdma_protocol_roce(dev, port_num)) 4529 return RDMA_AH_ATTR_TYPE_ROCE; 4530 if (rdma_protocol_ib(dev, port_num)) { 4531 if (rdma_cap_opa_ah(dev, port_num)) 4532 return RDMA_AH_ATTR_TYPE_OPA; 4533 return RDMA_AH_ATTR_TYPE_IB; 4534 } 4535 4536 return RDMA_AH_ATTR_TYPE_UNDEFINED; 4537 } 4538 4539 /** 4540 * ib_lid_cpu16 - Return lid in 16bit CPU encoding. 4541 * In the current implementation the only way to get 4542 * get the 32bit lid is from other sources for OPA. 4543 * For IB, lids will always be 16bits so cast the 4544 * value accordingly. 4545 * 4546 * @lid: A 32bit LID 4547 */ 4548 static inline u16 ib_lid_cpu16(u32 lid) 4549 { 4550 WARN_ON_ONCE(lid & 0xFFFF0000); 4551 return (u16)lid; 4552 } 4553 4554 /** 4555 * ib_lid_be16 - Return lid in 16bit BE encoding. 4556 * 4557 * @lid: A 32bit LID 4558 */ 4559 static inline __be16 ib_lid_be16(u32 lid) 4560 { 4561 WARN_ON_ONCE(lid & 0xFFFF0000); 4562 return cpu_to_be16((u16)lid); 4563 } 4564 4565 /** 4566 * ib_get_vector_affinity - Get the affinity mappings of a given completion 4567 * vector 4568 * @device: the rdma device 4569 * @comp_vector: index of completion vector 4570 * 4571 * Returns NULL on failure, otherwise a corresponding cpu map of the 4572 * completion vector (returns all-cpus map if the device driver doesn't 4573 * implement get_vector_affinity). 4574 */ 4575 static inline const struct cpumask * 4576 ib_get_vector_affinity(struct ib_device *device, int comp_vector) 4577 { 4578 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors || 4579 !device->ops.get_vector_affinity) 4580 return NULL; 4581 4582 return device->ops.get_vector_affinity(device, comp_vector); 4583 4584 } 4585 4586 /** 4587 * rdma_roce_rescan_device - Rescan all of the network devices in the system 4588 * and add their gids, as needed, to the relevant RoCE devices. 4589 * 4590 * @device: the rdma device 4591 */ 4592 void rdma_roce_rescan_device(struct ib_device *ibdev); 4593 4594 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile); 4595 4596 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs); 4597 4598 struct net_device *rdma_alloc_netdev(struct ib_device *device, u8 port_num, 4599 enum rdma_netdev_t type, const char *name, 4600 unsigned char name_assign_type, 4601 void (*setup)(struct net_device *)); 4602 4603 int rdma_init_netdev(struct ib_device *device, u8 port_num, 4604 enum rdma_netdev_t type, const char *name, 4605 unsigned char name_assign_type, 4606 void (*setup)(struct net_device *), 4607 struct net_device *netdev); 4608 4609 /** 4610 * rdma_set_device_sysfs_group - Set device attributes group to have 4611 * driver specific sysfs entries at 4612 * for infiniband class. 4613 * 4614 * @device: device pointer for which attributes to be created 4615 * @group: Pointer to group which should be added when device 4616 * is registered with sysfs. 4617 * rdma_set_device_sysfs_group() allows existing drivers to expose one 4618 * group per device to have sysfs attributes. 4619 * 4620 * NOTE: New drivers should not make use of this API; instead new device 4621 * parameter should be exposed via netlink command. This API and mechanism 4622 * exist only for existing drivers. 4623 */ 4624 static inline void 4625 rdma_set_device_sysfs_group(struct ib_device *dev, 4626 const struct attribute_group *group) 4627 { 4628 dev->groups[1] = group; 4629 } 4630 4631 /** 4632 * rdma_device_to_ibdev - Get ib_device pointer from device pointer 4633 * 4634 * @device: device pointer for which ib_device pointer to retrieve 4635 * 4636 * rdma_device_to_ibdev() retrieves ib_device pointer from device. 4637 * 4638 */ 4639 static inline struct ib_device *rdma_device_to_ibdev(struct device *device) 4640 { 4641 struct ib_core_device *coredev = 4642 container_of(device, struct ib_core_device, dev); 4643 4644 return coredev->owner; 4645 } 4646 4647 /** 4648 * rdma_device_to_drv_device - Helper macro to reach back to driver's 4649 * ib_device holder structure from device pointer. 4650 * 4651 * NOTE: New drivers should not make use of this API; This API is only for 4652 * existing drivers who have exposed sysfs entries using 4653 * rdma_set_device_sysfs_group(). 4654 */ 4655 #define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member) \ 4656 container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member) 4657 4658 bool rdma_dev_access_netns(const struct ib_device *device, 4659 const struct net *net); 4660 #endif /* IB_VERBS_H */ 4661