1 /* SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB */ 2 /* 3 * Copyright (c) 2013-2020, Mellanox Technologies inc. All rights reserved. 4 * Copyright (c) 2020, Intel Corporation. All rights reserved. 5 */ 6 7 #ifndef MLX5_IB_H 8 #define MLX5_IB_H 9 10 #include <linux/kernel.h> 11 #include <linux/sched.h> 12 #include <rdma/ib_verbs.h> 13 #include <rdma/ib_umem.h> 14 #include <rdma/ib_smi.h> 15 #include <linux/mlx5/driver.h> 16 #include <linux/mlx5/cq.h> 17 #include <linux/mlx5/fs.h> 18 #include <linux/mlx5/qp.h> 19 #include <linux/types.h> 20 #include <linux/mlx5/transobj.h> 21 #include <rdma/ib_user_verbs.h> 22 #include <rdma/mlx5-abi.h> 23 #include <rdma/uverbs_ioctl.h> 24 #include <rdma/mlx5_user_ioctl_cmds.h> 25 #include <rdma/mlx5_user_ioctl_verbs.h> 26 27 #include "srq.h" 28 #include "qp.h" 29 #include "macsec.h" 30 31 #define mlx5_ib_dbg(_dev, format, arg...) \ 32 dev_dbg(&(_dev)->ib_dev.dev, "%s:%d:(pid %d): " format, __func__, \ 33 __LINE__, current->pid, ##arg) 34 35 #define mlx5_ib_err(_dev, format, arg...) \ 36 dev_err(&(_dev)->ib_dev.dev, "%s:%d:(pid %d): " format, __func__, \ 37 __LINE__, current->pid, ##arg) 38 39 #define mlx5_ib_warn(_dev, format, arg...) \ 40 dev_warn(&(_dev)->ib_dev.dev, "%s:%d:(pid %d): " format, __func__, \ 41 __LINE__, current->pid, ##arg) 42 43 #define mlx5_ib_log(lvl, _dev, format, arg...) \ 44 dev_printk(lvl, &(_dev)->ib_dev.dev, "%s:%d:(pid %d): " format, \ 45 __func__, __LINE__, current->pid, ##arg) 46 47 #define MLX5_IB_DEFAULT_UIDX 0xffffff 48 #define MLX5_USER_ASSIGNED_UIDX_MASK __mlx5_mask(qpc, user_index) 49 50 static __always_inline unsigned long 51 __mlx5_log_page_size_to_bitmap(unsigned int log_pgsz_bits, 52 unsigned int pgsz_shift) 53 { 54 unsigned int largest_pg_shift = 55 min_t(unsigned long, (1ULL << log_pgsz_bits) - 1 + pgsz_shift, 56 BITS_PER_LONG - 1); 57 58 /* 59 * Despite a command allowing it, the device does not support lower than 60 * 4k page size. 61 */ 62 pgsz_shift = max_t(unsigned int, MLX5_ADAPTER_PAGE_SHIFT, pgsz_shift); 63 return GENMASK(largest_pg_shift, pgsz_shift); 64 } 65 66 /* 67 * For mkc users, instead of a page_offset the command has a start_iova which 68 * specifies both the page_offset and the on-the-wire IOVA 69 */ 70 #define mlx5_umem_find_best_pgsz(umem, typ, log_pgsz_fld, pgsz_shift, iova) \ 71 ib_umem_find_best_pgsz(umem, \ 72 __mlx5_log_page_size_to_bitmap( \ 73 __mlx5_bit_sz(typ, log_pgsz_fld), \ 74 pgsz_shift), \ 75 iova) 76 77 static __always_inline unsigned long 78 __mlx5_page_offset_to_bitmask(unsigned int page_offset_bits, 79 unsigned int offset_shift) 80 { 81 unsigned int largest_offset_shift = 82 min_t(unsigned long, page_offset_bits - 1 + offset_shift, 83 BITS_PER_LONG - 1); 84 85 return GENMASK(largest_offset_shift, offset_shift); 86 } 87 88 /* 89 * QP/CQ/WQ/etc type commands take a page offset that satisifies: 90 * page_offset_quantized * (page_size/scale) = page_offset 91 * Which restricts allowed page sizes to ones that satisify the above. 92 */ 93 unsigned long __mlx5_umem_find_best_quantized_pgoff( 94 struct ib_umem *umem, unsigned long pgsz_bitmap, 95 unsigned int page_offset_bits, u64 pgoff_bitmask, unsigned int scale, 96 unsigned int *page_offset_quantized); 97 #define mlx5_umem_find_best_quantized_pgoff(umem, typ, log_pgsz_fld, \ 98 pgsz_shift, page_offset_fld, \ 99 scale, page_offset_quantized) \ 100 __mlx5_umem_find_best_quantized_pgoff( \ 101 umem, \ 102 __mlx5_log_page_size_to_bitmap( \ 103 __mlx5_bit_sz(typ, log_pgsz_fld), pgsz_shift), \ 104 __mlx5_bit_sz(typ, page_offset_fld), \ 105 GENMASK(31, order_base_2(scale)), scale, \ 106 page_offset_quantized) 107 108 #define mlx5_umem_find_best_cq_quantized_pgoff(umem, typ, log_pgsz_fld, \ 109 pgsz_shift, page_offset_fld, \ 110 scale, page_offset_quantized) \ 111 __mlx5_umem_find_best_quantized_pgoff( \ 112 umem, \ 113 __mlx5_log_page_size_to_bitmap( \ 114 __mlx5_bit_sz(typ, log_pgsz_fld), pgsz_shift), \ 115 __mlx5_bit_sz(typ, page_offset_fld), 0, scale, \ 116 page_offset_quantized) 117 118 static inline unsigned long 119 mlx5_umem_dmabuf_find_best_pgsz(struct ib_umem_dmabuf *umem_dmabuf) 120 { 121 /* 122 * mkeys used for dmabuf are fixed at PAGE_SIZE because we must be able 123 * to hold any sgl after a move operation. Ideally the mkc page size 124 * could be changed at runtime to be optimal, but right now the driver 125 * cannot do that. 126 */ 127 return ib_umem_find_best_pgsz(&umem_dmabuf->umem, PAGE_SIZE, 128 umem_dmabuf->umem.iova); 129 } 130 131 enum { 132 MLX5_IB_MMAP_OFFSET_START = 9, 133 MLX5_IB_MMAP_OFFSET_END = 255, 134 }; 135 136 enum { 137 MLX5_IB_MMAP_CMD_SHIFT = 8, 138 MLX5_IB_MMAP_CMD_MASK = 0xff, 139 }; 140 141 enum { 142 MLX5_RES_SCAT_DATA32_CQE = 0x1, 143 MLX5_RES_SCAT_DATA64_CQE = 0x2, 144 MLX5_REQ_SCAT_DATA32_CQE = 0x11, 145 MLX5_REQ_SCAT_DATA64_CQE = 0x22, 146 }; 147 148 enum mlx5_ib_mad_ifc_flags { 149 MLX5_MAD_IFC_IGNORE_MKEY = 1, 150 MLX5_MAD_IFC_IGNORE_BKEY = 2, 151 MLX5_MAD_IFC_NET_VIEW = 4, 152 }; 153 154 enum { 155 MLX5_CROSS_CHANNEL_BFREG = 0, 156 }; 157 158 enum { 159 MLX5_CQE_VERSION_V0, 160 MLX5_CQE_VERSION_V1, 161 }; 162 163 enum { 164 MLX5_TM_MAX_RNDV_MSG_SIZE = 64, 165 MLX5_TM_MAX_SGE = 1, 166 }; 167 168 enum { 169 MLX5_IB_INVALID_UAR_INDEX = BIT(31), 170 MLX5_IB_INVALID_BFREG = BIT(31), 171 }; 172 173 enum { 174 MLX5_MAX_MEMIC_PAGES = 0x100, 175 MLX5_MEMIC_ALLOC_SIZE_MASK = 0x3f, 176 }; 177 178 enum { 179 MLX5_MEMIC_BASE_ALIGN = 6, 180 MLX5_MEMIC_BASE_SIZE = 1 << MLX5_MEMIC_BASE_ALIGN, 181 }; 182 183 enum mlx5_ib_mmap_type { 184 MLX5_IB_MMAP_TYPE_MEMIC = 1, 185 MLX5_IB_MMAP_TYPE_VAR = 2, 186 MLX5_IB_MMAP_TYPE_UAR_WC = 3, 187 MLX5_IB_MMAP_TYPE_UAR_NC = 4, 188 MLX5_IB_MMAP_TYPE_MEMIC_OP = 5, 189 }; 190 191 struct mlx5_bfreg_info { 192 u32 *sys_pages; 193 int num_low_latency_bfregs; 194 unsigned int *count; 195 196 /* 197 * protect bfreg allocation data structs 198 */ 199 struct mutex lock; 200 u32 ver; 201 u8 lib_uar_4k : 1; 202 u8 lib_uar_dyn : 1; 203 u32 num_sys_pages; 204 u32 num_static_sys_pages; 205 u32 total_num_bfregs; 206 u32 num_dyn_bfregs; 207 }; 208 209 struct mlx5_ib_ucontext { 210 struct ib_ucontext ibucontext; 211 struct list_head db_page_list; 212 213 /* protect doorbell record alloc/free 214 */ 215 struct mutex db_page_mutex; 216 struct mlx5_bfreg_info bfregi; 217 u8 cqe_version; 218 /* Transport Domain number */ 219 u32 tdn; 220 221 u64 lib_caps; 222 u16 devx_uid; 223 /* For RoCE LAG TX affinity */ 224 atomic_t tx_port_affinity; 225 }; 226 227 static inline struct mlx5_ib_ucontext *to_mucontext(struct ib_ucontext *ibucontext) 228 { 229 return container_of(ibucontext, struct mlx5_ib_ucontext, ibucontext); 230 } 231 232 struct mlx5_ib_pd { 233 struct ib_pd ibpd; 234 u32 pdn; 235 u16 uid; 236 }; 237 238 enum { 239 MLX5_IB_FLOW_ACTION_MODIFY_HEADER, 240 MLX5_IB_FLOW_ACTION_PACKET_REFORMAT, 241 MLX5_IB_FLOW_ACTION_DECAP, 242 }; 243 244 #define MLX5_IB_FLOW_MCAST_PRIO (MLX5_BY_PASS_NUM_PRIOS - 1) 245 #define MLX5_IB_FLOW_LAST_PRIO (MLX5_BY_PASS_NUM_REGULAR_PRIOS - 1) 246 #if (MLX5_IB_FLOW_LAST_PRIO <= 0) 247 #error "Invalid number of bypass priorities" 248 #endif 249 #define MLX5_IB_FLOW_LEFTOVERS_PRIO (MLX5_IB_FLOW_MCAST_PRIO + 1) 250 251 #define MLX5_IB_NUM_FLOW_FT (MLX5_IB_FLOW_LEFTOVERS_PRIO + 1) 252 #define MLX5_IB_NUM_SNIFFER_FTS 2 253 #define MLX5_IB_NUM_EGRESS_FTS 1 254 #define MLX5_IB_NUM_FDB_FTS MLX5_BY_PASS_NUM_REGULAR_PRIOS 255 256 struct mlx5_ib_anchor { 257 struct mlx5_flow_table *ft; 258 struct mlx5_flow_group *fg_goto_table; 259 struct mlx5_flow_group *fg_drop; 260 struct mlx5_flow_handle *rule_goto_table; 261 struct mlx5_flow_handle *rule_drop; 262 unsigned int rule_goto_table_ref; 263 }; 264 265 struct mlx5_ib_flow_prio { 266 struct mlx5_flow_table *flow_table; 267 struct mlx5_ib_anchor anchor; 268 unsigned int refcount; 269 }; 270 271 struct mlx5_ib_flow_handler { 272 struct list_head list; 273 struct ib_flow ibflow; 274 struct mlx5_ib_flow_prio *prio; 275 struct mlx5_flow_handle *rule; 276 struct ib_counters *ibcounters; 277 struct mlx5_ib_dev *dev; 278 struct mlx5_ib_flow_matcher *flow_matcher; 279 }; 280 281 struct mlx5_ib_flow_matcher { 282 struct mlx5_ib_match_params matcher_mask; 283 int mask_len; 284 enum mlx5_ib_flow_type flow_type; 285 enum mlx5_flow_namespace_type ns_type; 286 u16 priority; 287 struct mlx5_core_dev *mdev; 288 atomic_t usecnt; 289 u8 match_criteria_enable; 290 }; 291 292 struct mlx5_ib_steering_anchor { 293 struct mlx5_ib_flow_prio *ft_prio; 294 struct mlx5_ib_dev *dev; 295 atomic_t usecnt; 296 }; 297 298 struct mlx5_ib_pp { 299 u16 index; 300 struct mlx5_core_dev *mdev; 301 }; 302 303 enum mlx5_ib_optional_counter_type { 304 MLX5_IB_OPCOUNTER_CC_RX_CE_PKTS, 305 MLX5_IB_OPCOUNTER_CC_RX_CNP_PKTS, 306 MLX5_IB_OPCOUNTER_CC_TX_CNP_PKTS, 307 308 MLX5_IB_OPCOUNTER_MAX, 309 }; 310 311 struct mlx5_ib_flow_db { 312 struct mlx5_ib_flow_prio prios[MLX5_IB_NUM_FLOW_FT]; 313 struct mlx5_ib_flow_prio egress_prios[MLX5_IB_NUM_FLOW_FT]; 314 struct mlx5_ib_flow_prio sniffer[MLX5_IB_NUM_SNIFFER_FTS]; 315 struct mlx5_ib_flow_prio egress[MLX5_IB_NUM_EGRESS_FTS]; 316 struct mlx5_ib_flow_prio fdb[MLX5_IB_NUM_FDB_FTS]; 317 struct mlx5_ib_flow_prio rdma_rx[MLX5_IB_NUM_FLOW_FT]; 318 struct mlx5_ib_flow_prio rdma_tx[MLX5_IB_NUM_FLOW_FT]; 319 struct mlx5_ib_flow_prio opfcs[MLX5_IB_OPCOUNTER_MAX]; 320 struct mlx5_flow_table *lag_demux_ft; 321 /* Protect flow steering bypass flow tables 322 * when add/del flow rules. 323 * only single add/removal of flow steering rule could be done 324 * simultaneously. 325 */ 326 struct mutex lock; 327 }; 328 329 /* Use macros here so that don't have to duplicate 330 * enum ib_qp_type for low-level driver 331 */ 332 333 #define MLX5_IB_QPT_REG_UMR IB_QPT_RESERVED1 334 /* 335 * IB_QPT_GSI creates the software wrapper around GSI, and MLX5_IB_QPT_HW_GSI 336 * creates the actual hardware QP. 337 */ 338 #define MLX5_IB_QPT_HW_GSI IB_QPT_RESERVED2 339 #define MLX5_IB_QPT_DCI IB_QPT_RESERVED3 340 #define MLX5_IB_QPT_DCT IB_QPT_RESERVED4 341 #define MLX5_IB_WR_UMR IB_WR_RESERVED1 342 343 #define MLX5_IB_UPD_XLT_ZAP BIT(0) 344 #define MLX5_IB_UPD_XLT_ENABLE BIT(1) 345 #define MLX5_IB_UPD_XLT_ATOMIC BIT(2) 346 #define MLX5_IB_UPD_XLT_ADDR BIT(3) 347 #define MLX5_IB_UPD_XLT_PD BIT(4) 348 #define MLX5_IB_UPD_XLT_ACCESS BIT(5) 349 #define MLX5_IB_UPD_XLT_INDIRECT BIT(6) 350 351 /* Private QP creation flags to be passed in ib_qp_init_attr.create_flags. 352 * 353 * These flags are intended for internal use by the mlx5_ib driver, and they 354 * rely on the range reserved for that use in the ib_qp_create_flags enum. 355 */ 356 #define MLX5_IB_QP_CREATE_SQPN_QP1 IB_QP_CREATE_RESERVED_START 357 #define MLX5_IB_QP_CREATE_WC_TEST (IB_QP_CREATE_RESERVED_START << 1) 358 359 struct wr_list { 360 u16 opcode; 361 u16 next; 362 }; 363 364 enum mlx5_ib_rq_flags { 365 MLX5_IB_RQ_CVLAN_STRIPPING = 1 << 0, 366 MLX5_IB_RQ_PCI_WRITE_END_PADDING = 1 << 1, 367 }; 368 369 struct mlx5_ib_wq { 370 struct mlx5_frag_buf_ctrl fbc; 371 u64 *wrid; 372 u32 *wr_data; 373 struct wr_list *w_list; 374 unsigned *wqe_head; 375 u16 unsig_count; 376 377 /* serialize post to the work queue 378 */ 379 spinlock_t lock; 380 int wqe_cnt; 381 int max_post; 382 int max_gs; 383 int offset; 384 int wqe_shift; 385 unsigned head; 386 unsigned tail; 387 u16 cur_post; 388 u16 last_poll; 389 void *cur_edge; 390 }; 391 392 enum mlx5_ib_wq_flags { 393 MLX5_IB_WQ_FLAGS_DELAY_DROP = 0x1, 394 MLX5_IB_WQ_FLAGS_STRIDING_RQ = 0x2, 395 }; 396 397 #define MLX5_MIN_SINGLE_WQE_LOG_NUM_STRIDES 9 398 #define MLX5_MAX_SINGLE_WQE_LOG_NUM_STRIDES 16 399 #define MLX5_MIN_SINGLE_STRIDE_LOG_NUM_BYTES 6 400 #define MLX5_MAX_SINGLE_STRIDE_LOG_NUM_BYTES 13 401 #define MLX5_EXT_MIN_SINGLE_WQE_LOG_NUM_STRIDES 3 402 403 struct mlx5_ib_rwq { 404 struct ib_wq ibwq; 405 struct mlx5_core_qp core_qp; 406 u32 rq_num_pas; 407 u32 log_rq_stride; 408 u32 log_rq_size; 409 u32 rq_page_offset; 410 u32 log_page_size; 411 u32 log_num_strides; 412 u32 two_byte_shift_en; 413 u32 single_stride_log_num_of_bytes; 414 struct ib_umem *umem; 415 size_t buf_size; 416 unsigned int page_shift; 417 struct mlx5_db db; 418 u32 user_index; 419 u32 wqe_count; 420 u32 wqe_shift; 421 int wq_sig; 422 u32 create_flags; /* Use enum mlx5_ib_wq_flags */ 423 }; 424 425 struct mlx5_ib_rwq_ind_table { 426 struct ib_rwq_ind_table ib_rwq_ind_tbl; 427 u32 rqtn; 428 u16 uid; 429 }; 430 431 struct mlx5_ib_ubuffer { 432 struct ib_umem *umem; 433 int buf_size; 434 u64 buf_addr; 435 }; 436 437 struct mlx5_ib_qp_base { 438 struct mlx5_ib_qp *container_mibqp; 439 struct mlx5_core_qp mqp; 440 struct mlx5_ib_ubuffer ubuffer; 441 }; 442 443 struct mlx5_ib_qp_trans { 444 struct mlx5_ib_qp_base base; 445 u16 xrcdn; 446 u32 alt_port; 447 u8 atomic_rd_en; 448 u8 resp_depth; 449 }; 450 451 struct mlx5_ib_rss_qp { 452 u32 tirn; 453 }; 454 455 struct mlx5_ib_rq { 456 struct mlx5_ib_qp_base base; 457 struct mlx5_ib_wq *rq; 458 struct mlx5_ib_ubuffer ubuffer; 459 struct mlx5_db *doorbell; 460 u32 tirn; 461 u8 state; 462 u32 flags; 463 }; 464 465 struct mlx5_ib_sq { 466 struct mlx5_ib_qp_base base; 467 struct mlx5_ib_wq *sq; 468 struct mlx5_ib_ubuffer ubuffer; 469 struct mlx5_db *doorbell; 470 struct mlx5_flow_handle *flow_rule; 471 u32 tisn; 472 u8 state; 473 }; 474 475 struct mlx5_ib_raw_packet_qp { 476 struct mlx5_ib_sq sq; 477 struct mlx5_ib_rq rq; 478 }; 479 480 struct mlx5_bf { 481 int buf_size; 482 unsigned long offset; 483 struct mlx5_sq_bfreg *bfreg; 484 }; 485 486 struct mlx5_ib_dct { 487 struct mlx5_core_dct mdct; 488 u32 *in; 489 }; 490 491 struct mlx5_ib_gsi_qp { 492 struct ib_qp *rx_qp; 493 u32 port_num; 494 struct ib_qp_cap cap; 495 struct ib_cq *cq; 496 struct mlx5_ib_gsi_wr *outstanding_wrs; 497 u32 outstanding_pi, outstanding_ci; 498 int num_qps; 499 /* Protects access to the tx_qps. Post send operations synchronize 500 * with tx_qp creation in setup_qp(). Also protects the 501 * outstanding_wrs array and indices. 502 */ 503 spinlock_t lock; 504 struct ib_qp **tx_qps; 505 }; 506 507 struct mlx5_ib_qp { 508 struct ib_qp ibqp; 509 union { 510 struct mlx5_ib_qp_trans trans_qp; 511 struct mlx5_ib_raw_packet_qp raw_packet_qp; 512 struct mlx5_ib_rss_qp rss_qp; 513 struct mlx5_ib_dct dct; 514 struct mlx5_ib_gsi_qp gsi; 515 }; 516 struct mlx5_frag_buf buf; 517 518 struct mlx5_db db; 519 struct mlx5_ib_wq rq; 520 521 u8 sq_signal_bits; 522 u8 next_fence; 523 struct mlx5_ib_wq sq; 524 525 /* serialize qp state modifications 526 */ 527 struct mutex mutex; 528 /* cached variant of create_flags from struct ib_qp_init_attr */ 529 u32 flags; 530 u32 port; 531 u8 state; 532 int max_inline_data; 533 struct mlx5_bf bf; 534 u8 has_rq:1; 535 u8 is_rss:1; 536 537 /* only for user space QPs. For kernel 538 * we have it from the bf object 539 */ 540 int bfregn; 541 542 struct list_head qps_list; 543 struct list_head cq_recv_list; 544 struct list_head cq_send_list; 545 struct mlx5_rate_limit rl; 546 u32 underlay_qpn; 547 u32 flags_en; 548 /* 549 * IB/core doesn't store low-level QP types, so 550 * store both MLX and IBTA types in the field below. 551 */ 552 enum ib_qp_type type; 553 /* A flag to indicate if there's a new counter is configured 554 * but not take effective 555 */ 556 u32 counter_pending; 557 u16 gsi_lag_port; 558 }; 559 560 struct mlx5_ib_cq_buf { 561 struct mlx5_frag_buf_ctrl fbc; 562 struct mlx5_frag_buf frag_buf; 563 struct ib_umem *umem; 564 int cqe_size; 565 int nent; 566 }; 567 568 enum mlx5_ib_cq_pr_flags { 569 MLX5_IB_CQ_PR_FLAGS_CQE_128_PAD = 1 << 0, 570 MLX5_IB_CQ_PR_FLAGS_REAL_TIME_TS = 1 << 1, 571 }; 572 573 struct mlx5_ib_cq { 574 struct ib_cq ibcq; 575 struct mlx5_core_cq mcq; 576 struct mlx5_ib_cq_buf buf; 577 struct mlx5_db db; 578 579 /* serialize access to the CQ 580 */ 581 spinlock_t lock; 582 583 /* protect resize cq 584 */ 585 struct mutex resize_mutex; 586 struct mlx5_ib_cq_buf *resize_buf; 587 struct ib_umem *resize_umem; 588 int cqe_size; 589 struct list_head list_send_qp; 590 struct list_head list_recv_qp; 591 u32 create_flags; 592 struct list_head wc_list; 593 enum ib_cq_notify_flags notify_flags; 594 struct work_struct notify_work; 595 u16 private_flags; /* Use mlx5_ib_cq_pr_flags */ 596 }; 597 598 struct mlx5_ib_wc { 599 struct ib_wc wc; 600 struct list_head list; 601 }; 602 603 struct mlx5_ib_srq { 604 struct ib_srq ibsrq; 605 struct mlx5_core_srq msrq; 606 struct mlx5_frag_buf buf; 607 struct mlx5_db db; 608 struct mlx5_frag_buf_ctrl fbc; 609 u64 *wrid; 610 /* protect SRQ hanlding 611 */ 612 spinlock_t lock; 613 int head; 614 int tail; 615 u16 wqe_ctr; 616 struct ib_umem *umem; 617 /* serialize arming a SRQ 618 */ 619 struct mutex mutex; 620 int wq_sig; 621 }; 622 623 struct mlx5_ib_xrcd { 624 struct ib_xrcd ibxrcd; 625 u32 xrcdn; 626 }; 627 628 enum mlx5_ib_mtt_access_flags { 629 MLX5_IB_MTT_READ = (1 << 0), 630 MLX5_IB_MTT_WRITE = (1 << 1), 631 }; 632 633 struct mlx5_user_mmap_entry { 634 struct rdma_user_mmap_entry rdma_entry; 635 u8 mmap_flag; 636 u64 address; 637 u32 page_idx; 638 }; 639 640 enum mlx5_mkey_type { 641 MLX5_MKEY_MR = 1, 642 MLX5_MKEY_MW, 643 MLX5_MKEY_INDIRECT_DEVX, 644 }; 645 646 struct mlx5r_cache_rb_key { 647 u8 ats:1; 648 unsigned int access_mode; 649 unsigned int access_flags; 650 unsigned int ndescs; 651 }; 652 653 struct mlx5_ib_mkey { 654 u32 key; 655 enum mlx5_mkey_type type; 656 unsigned int ndescs; 657 struct wait_queue_head wait; 658 refcount_t usecount; 659 /* Cacheable user Mkey must hold either a rb_key or a cache_ent. */ 660 struct mlx5r_cache_rb_key rb_key; 661 struct mlx5_cache_ent *cache_ent; 662 }; 663 664 #define MLX5_IB_MTT_PRESENT (MLX5_IB_MTT_READ | MLX5_IB_MTT_WRITE) 665 666 #define MLX5_IB_DM_MEMIC_ALLOWED_ACCESS (IB_ACCESS_LOCAL_WRITE |\ 667 IB_ACCESS_REMOTE_WRITE |\ 668 IB_ACCESS_REMOTE_READ |\ 669 IB_ACCESS_REMOTE_ATOMIC |\ 670 IB_ZERO_BASED) 671 672 #define MLX5_IB_DM_SW_ICM_ALLOWED_ACCESS (IB_ACCESS_LOCAL_WRITE |\ 673 IB_ACCESS_REMOTE_WRITE |\ 674 IB_ACCESS_REMOTE_READ |\ 675 IB_ZERO_BASED) 676 677 #define mlx5_update_odp_stats(mr, counter_name, value) \ 678 atomic64_add(value, &((mr)->odp_stats.counter_name)) 679 680 struct mlx5_ib_mr { 681 struct ib_mr ibmr; 682 struct mlx5_ib_mkey mmkey; 683 684 struct ib_umem *umem; 685 686 union { 687 /* Used only by kernel MRs (umem == NULL) */ 688 struct { 689 void *descs; 690 void *descs_alloc; 691 dma_addr_t desc_map; 692 int max_descs; 693 int desc_size; 694 int access_mode; 695 696 /* For Kernel IB_MR_TYPE_INTEGRITY */ 697 struct mlx5_core_sig_ctx *sig; 698 struct mlx5_ib_mr *pi_mr; 699 struct mlx5_ib_mr *klm_mr; 700 struct mlx5_ib_mr *mtt_mr; 701 u64 data_iova; 702 u64 pi_iova; 703 int meta_ndescs; 704 int meta_length; 705 int data_length; 706 }; 707 708 /* Used only by User MRs (umem != NULL) */ 709 struct { 710 unsigned int page_shift; 711 /* Current access_flags */ 712 int access_flags; 713 714 /* For User ODP */ 715 struct mlx5_ib_mr *parent; 716 struct xarray implicit_children; 717 union { 718 struct work_struct work; 719 } odp_destroy; 720 struct ib_odp_counters odp_stats; 721 bool is_odp_implicit; 722 }; 723 }; 724 }; 725 726 static inline bool is_odp_mr(struct mlx5_ib_mr *mr) 727 { 728 return IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING) && mr->umem && 729 mr->umem->is_odp; 730 } 731 732 static inline bool is_dmabuf_mr(struct mlx5_ib_mr *mr) 733 { 734 return IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING) && mr->umem && 735 mr->umem->is_dmabuf; 736 } 737 738 struct mlx5_ib_mw { 739 struct ib_mw ibmw; 740 struct mlx5_ib_mkey mmkey; 741 }; 742 743 struct mlx5_ib_umr_context { 744 struct ib_cqe cqe; 745 enum ib_wc_status status; 746 struct completion done; 747 }; 748 749 enum { 750 MLX5_UMR_STATE_UNINIT, 751 MLX5_UMR_STATE_ACTIVE, 752 MLX5_UMR_STATE_RECOVER, 753 MLX5_UMR_STATE_ERR, 754 }; 755 756 struct umr_common { 757 struct ib_pd *pd; 758 struct ib_cq *cq; 759 struct ib_qp *qp; 760 /* Protects from UMR QP overflow 761 */ 762 struct semaphore sem; 763 /* Protects from using UMR while the UMR is not active 764 */ 765 struct mutex lock; 766 unsigned int state; 767 }; 768 769 struct mlx5_cache_ent { 770 struct xarray mkeys; 771 unsigned long stored; 772 unsigned long reserved; 773 774 char name[4]; 775 776 struct rb_node node; 777 struct mlx5r_cache_rb_key rb_key; 778 779 u8 is_tmp:1; 780 u8 disabled:1; 781 u8 fill_to_high_water:1; 782 783 /* 784 * - limit is the low water mark for stored mkeys, 2* limit is the 785 * upper water mark. 786 */ 787 u32 in_use; 788 u32 limit; 789 790 /* Statistics */ 791 u32 miss; 792 793 struct mlx5_ib_dev *dev; 794 struct delayed_work dwork; 795 }; 796 797 struct mlx5r_async_create_mkey { 798 union { 799 u32 in[MLX5_ST_SZ_BYTES(create_mkey_in)]; 800 u32 out[MLX5_ST_SZ_DW(create_mkey_out)]; 801 }; 802 struct mlx5_async_work cb_work; 803 struct mlx5_cache_ent *ent; 804 u32 mkey; 805 }; 806 807 struct mlx5_mkey_cache { 808 struct workqueue_struct *wq; 809 struct rb_root rb_root; 810 struct mutex rb_lock; 811 struct dentry *fs_root; 812 unsigned long last_add; 813 struct delayed_work remove_ent_dwork; 814 }; 815 816 struct mlx5_ib_port_resources { 817 struct mlx5_ib_gsi_qp *gsi; 818 struct work_struct pkey_change_work; 819 }; 820 821 struct mlx5_ib_resources { 822 struct ib_cq *c0; 823 struct mutex cq_lock; 824 u32 xrcdn0; 825 u32 xrcdn1; 826 struct ib_pd *p0; 827 struct ib_srq *s0; 828 struct ib_srq *s1; 829 struct mutex srq_lock; 830 struct mlx5_ib_port_resources ports[2]; 831 }; 832 833 #define MAX_OPFC_RULES 2 834 835 struct mlx5_ib_op_fc { 836 struct mlx5_fc *fc; 837 struct mlx5_flow_handle *rule[MAX_OPFC_RULES]; 838 }; 839 840 struct mlx5_ib_counters { 841 struct rdma_stat_desc *descs; 842 size_t *offsets; 843 u32 num_q_counters; 844 u32 num_cong_counters; 845 u32 num_ext_ppcnt_counters; 846 u32 num_op_counters; 847 u16 set_id; 848 struct mlx5_ib_op_fc opfcs[MLX5_IB_OPCOUNTER_MAX]; 849 }; 850 851 int mlx5_ib_fs_add_op_fc(struct mlx5_ib_dev *dev, u32 port_num, 852 struct mlx5_ib_op_fc *opfc, 853 enum mlx5_ib_optional_counter_type type); 854 855 void mlx5_ib_fs_remove_op_fc(struct mlx5_ib_dev *dev, 856 struct mlx5_ib_op_fc *opfc, 857 enum mlx5_ib_optional_counter_type type); 858 859 struct mlx5_ib_multiport_info; 860 861 struct mlx5_ib_multiport { 862 struct mlx5_ib_multiport_info *mpi; 863 /* To be held when accessing the multiport info */ 864 spinlock_t mpi_lock; 865 }; 866 867 struct mlx5_roce { 868 /* Protect mlx5_ib_get_netdev from invoking dev_hold() with a NULL 869 * netdev pointer 870 */ 871 rwlock_t netdev_lock; 872 struct net_device *netdev; 873 struct notifier_block nb; 874 struct netdev_net_notifier nn; 875 struct notifier_block mdev_nb; 876 struct net_device *tracking_netdev; 877 atomic_t tx_port_affinity; 878 enum ib_port_state last_port_state; 879 struct mlx5_ib_dev *dev; 880 u32 native_port_num; 881 }; 882 883 struct mlx5_ib_port { 884 struct mlx5_ib_counters cnts; 885 struct mlx5_ib_multiport mp; 886 struct mlx5_ib_dbg_cc_params *dbg_cc_params; 887 struct mlx5_roce roce; 888 struct mlx5_eswitch_rep *rep; 889 #ifdef CONFIG_MLX5_MACSEC 890 struct mlx5_reserved_gids *reserved_gids; 891 #endif 892 }; 893 894 struct mlx5_ib_dbg_param { 895 int offset; 896 struct mlx5_ib_dev *dev; 897 struct dentry *dentry; 898 u32 port_num; 899 }; 900 901 enum mlx5_ib_dbg_cc_types { 902 MLX5_IB_DBG_CC_RP_CLAMP_TGT_RATE, 903 MLX5_IB_DBG_CC_RP_CLAMP_TGT_RATE_ATI, 904 MLX5_IB_DBG_CC_RP_TIME_RESET, 905 MLX5_IB_DBG_CC_RP_BYTE_RESET, 906 MLX5_IB_DBG_CC_RP_THRESHOLD, 907 MLX5_IB_DBG_CC_RP_AI_RATE, 908 MLX5_IB_DBG_CC_RP_MAX_RATE, 909 MLX5_IB_DBG_CC_RP_HAI_RATE, 910 MLX5_IB_DBG_CC_RP_MIN_DEC_FAC, 911 MLX5_IB_DBG_CC_RP_MIN_RATE, 912 MLX5_IB_DBG_CC_RP_RATE_TO_SET_ON_FIRST_CNP, 913 MLX5_IB_DBG_CC_RP_DCE_TCP_G, 914 MLX5_IB_DBG_CC_RP_DCE_TCP_RTT, 915 MLX5_IB_DBG_CC_RP_RATE_REDUCE_MONITOR_PERIOD, 916 MLX5_IB_DBG_CC_RP_INITIAL_ALPHA_VALUE, 917 MLX5_IB_DBG_CC_RP_GD, 918 MLX5_IB_DBG_CC_NP_MIN_TIME_BETWEEN_CNPS, 919 MLX5_IB_DBG_CC_NP_CNP_DSCP, 920 MLX5_IB_DBG_CC_NP_CNP_PRIO_MODE, 921 MLX5_IB_DBG_CC_NP_CNP_PRIO, 922 MLX5_IB_DBG_CC_GENERAL_RTT_RESP_DSCP_VALID, 923 MLX5_IB_DBG_CC_GENERAL_RTT_RESP_DSCP, 924 MLX5_IB_DBG_CC_MAX, 925 }; 926 927 struct mlx5_ib_dbg_cc_params { 928 struct dentry *root; 929 struct mlx5_ib_dbg_param params[MLX5_IB_DBG_CC_MAX]; 930 }; 931 932 enum { 933 MLX5_MAX_DELAY_DROP_TIMEOUT_MS = 100, 934 }; 935 936 struct mlx5_ib_delay_drop { 937 struct mlx5_ib_dev *dev; 938 struct work_struct delay_drop_work; 939 /* serialize setting of delay drop */ 940 struct mutex lock; 941 u32 timeout; 942 bool activate; 943 atomic_t events_cnt; 944 atomic_t rqs_cnt; 945 struct dentry *dir_debugfs; 946 }; 947 948 enum mlx5_ib_stages { 949 MLX5_IB_STAGE_INIT, 950 MLX5_IB_STAGE_FS, 951 MLX5_IB_STAGE_CAPS, 952 MLX5_IB_STAGE_NON_DEFAULT_CB, 953 MLX5_IB_STAGE_ROCE, 954 MLX5_IB_STAGE_QP, 955 MLX5_IB_STAGE_SRQ, 956 MLX5_IB_STAGE_DEVICE_RESOURCES, 957 MLX5_IB_STAGE_ODP, 958 MLX5_IB_STAGE_COUNTERS, 959 MLX5_IB_STAGE_CONG_DEBUGFS, 960 MLX5_IB_STAGE_UAR, 961 MLX5_IB_STAGE_BFREG, 962 MLX5_IB_STAGE_PRE_IB_REG_UMR, 963 MLX5_IB_STAGE_WHITELIST_UID, 964 MLX5_IB_STAGE_IB_REG, 965 MLX5_IB_STAGE_DEVICE_NOTIFIER, 966 MLX5_IB_STAGE_POST_IB_REG_UMR, 967 MLX5_IB_STAGE_DELAY_DROP, 968 MLX5_IB_STAGE_RESTRACK, 969 MLX5_IB_STAGE_MAX, 970 }; 971 972 struct mlx5_ib_stage { 973 int (*init)(struct mlx5_ib_dev *dev); 974 void (*cleanup)(struct mlx5_ib_dev *dev); 975 }; 976 977 #define STAGE_CREATE(_stage, _init, _cleanup) \ 978 .stage[_stage] = {.init = _init, .cleanup = _cleanup} 979 980 struct mlx5_ib_profile { 981 struct mlx5_ib_stage stage[MLX5_IB_STAGE_MAX]; 982 }; 983 984 struct mlx5_ib_multiport_info { 985 struct list_head list; 986 struct mlx5_ib_dev *ibdev; 987 struct mlx5_core_dev *mdev; 988 struct notifier_block mdev_events; 989 struct completion unref_comp; 990 u64 sys_image_guid; 991 u32 mdev_refcnt; 992 bool is_master; 993 bool unaffiliate; 994 }; 995 996 struct mlx5_ib_flow_action { 997 struct ib_flow_action ib_action; 998 union { 999 struct { 1000 u64 ib_flags; 1001 struct mlx5_accel_esp_xfrm *ctx; 1002 } esp_aes_gcm; 1003 struct { 1004 struct mlx5_ib_dev *dev; 1005 u32 sub_type; 1006 union { 1007 struct mlx5_modify_hdr *modify_hdr; 1008 struct mlx5_pkt_reformat *pkt_reformat; 1009 }; 1010 } flow_action_raw; 1011 }; 1012 }; 1013 1014 struct mlx5_dm { 1015 struct mlx5_core_dev *dev; 1016 /* This lock is used to protect the access to the shared 1017 * allocation map when concurrent requests by different 1018 * processes are handled. 1019 */ 1020 spinlock_t lock; 1021 DECLARE_BITMAP(memic_alloc_pages, MLX5_MAX_MEMIC_PAGES); 1022 }; 1023 1024 struct mlx5_read_counters_attr { 1025 struct mlx5_fc *hw_cntrs_hndl; 1026 u64 *out; 1027 u32 flags; 1028 }; 1029 1030 enum mlx5_ib_counters_type { 1031 MLX5_IB_COUNTERS_FLOW, 1032 }; 1033 1034 struct mlx5_ib_mcounters { 1035 struct ib_counters ibcntrs; 1036 enum mlx5_ib_counters_type type; 1037 /* number of counters supported for this counters type */ 1038 u32 counters_num; 1039 struct mlx5_fc *hw_cntrs_hndl; 1040 /* read function for this counters type */ 1041 int (*read_counters)(struct ib_device *ibdev, 1042 struct mlx5_read_counters_attr *read_attr); 1043 /* max index set as part of create_flow */ 1044 u32 cntrs_max_index; 1045 /* number of counters data entries (<description,index> pair) */ 1046 u32 ncounters; 1047 /* counters data array for descriptions and indexes */ 1048 struct mlx5_ib_flow_counters_desc *counters_data; 1049 /* protects access to mcounters internal data */ 1050 struct mutex mcntrs_mutex; 1051 }; 1052 1053 static inline struct mlx5_ib_mcounters * 1054 to_mcounters(struct ib_counters *ibcntrs) 1055 { 1056 return container_of(ibcntrs, struct mlx5_ib_mcounters, ibcntrs); 1057 } 1058 1059 int parse_flow_flow_action(struct mlx5_ib_flow_action *maction, 1060 bool is_egress, 1061 struct mlx5_flow_act *action); 1062 struct mlx5_ib_lb_state { 1063 /* protect the user_td */ 1064 struct mutex mutex; 1065 u32 user_td; 1066 int qps; 1067 bool enabled; 1068 }; 1069 1070 struct mlx5_ib_pf_eq { 1071 struct notifier_block irq_nb; 1072 struct mlx5_ib_dev *dev; 1073 struct mlx5_eq *core; 1074 struct work_struct work; 1075 spinlock_t lock; /* Pagefaults spinlock */ 1076 struct workqueue_struct *wq; 1077 mempool_t *pool; 1078 }; 1079 1080 struct mlx5_devx_event_table { 1081 struct mlx5_nb devx_nb; 1082 /* serialize updating the event_xa */ 1083 struct mutex event_xa_lock; 1084 struct xarray event_xa; 1085 }; 1086 1087 struct mlx5_var_table { 1088 /* serialize updating the bitmap */ 1089 struct mutex bitmap_lock; 1090 unsigned long *bitmap; 1091 u64 hw_start_addr; 1092 u32 stride_size; 1093 u64 num_var_hw_entries; 1094 }; 1095 1096 struct mlx5_port_caps { 1097 bool has_smi; 1098 u8 ext_port_cap; 1099 }; 1100 1101 1102 struct mlx5_special_mkeys { 1103 u32 dump_fill_mkey; 1104 __be32 null_mkey; 1105 __be32 terminate_scatter_list_mkey; 1106 }; 1107 1108 struct mlx5_macsec { 1109 struct mutex lock; /* Protects mlx5_macsec internal contexts */ 1110 struct list_head macsec_devices_list; 1111 struct notifier_block blocking_events_nb; 1112 }; 1113 1114 struct mlx5_ib_dev { 1115 struct ib_device ib_dev; 1116 struct mlx5_core_dev *mdev; 1117 struct notifier_block mdev_events; 1118 int num_ports; 1119 /* serialize update of capability mask 1120 */ 1121 struct mutex cap_mask_mutex; 1122 u8 ib_active:1; 1123 u8 is_rep:1; 1124 u8 lag_active:1; 1125 u8 wc_support:1; 1126 u8 fill_delay; 1127 struct umr_common umrc; 1128 /* sync used page count stats 1129 */ 1130 struct mlx5_ib_resources devr; 1131 1132 atomic_t mkey_var; 1133 struct mlx5_mkey_cache cache; 1134 struct timer_list delay_timer; 1135 /* Prevents soft lock on massive reg MRs */ 1136 struct mutex slow_path_mutex; 1137 struct ib_odp_caps odp_caps; 1138 u64 odp_max_size; 1139 struct mutex odp_eq_mutex; 1140 struct mlx5_ib_pf_eq odp_pf_eq; 1141 1142 struct xarray odp_mkeys; 1143 1144 struct mlx5_ib_flow_db *flow_db; 1145 /* protect resources needed as part of reset flow */ 1146 spinlock_t reset_flow_resource_lock; 1147 struct list_head qp_list; 1148 /* Array with num_ports elements */ 1149 struct mlx5_ib_port *port; 1150 struct mlx5_sq_bfreg bfreg; 1151 struct mlx5_sq_bfreg wc_bfreg; 1152 struct mlx5_sq_bfreg fp_bfreg; 1153 struct mlx5_ib_delay_drop delay_drop; 1154 const struct mlx5_ib_profile *profile; 1155 1156 struct mlx5_ib_lb_state lb; 1157 u8 umr_fence; 1158 struct list_head ib_dev_list; 1159 u64 sys_image_guid; 1160 struct mlx5_dm dm; 1161 u16 devx_whitelist_uid; 1162 struct mlx5_srq_table srq_table; 1163 struct mlx5_qp_table qp_table; 1164 struct mlx5_async_ctx async_ctx; 1165 struct mlx5_devx_event_table devx_event_table; 1166 struct mlx5_var_table var_table; 1167 1168 struct xarray sig_mrs; 1169 struct mlx5_port_caps port_caps[MLX5_MAX_PORTS]; 1170 u16 pkey_table_len; 1171 u8 lag_ports; 1172 struct mlx5_special_mkeys mkeys; 1173 1174 #ifdef CONFIG_MLX5_MACSEC 1175 struct mlx5_macsec macsec; 1176 #endif 1177 }; 1178 1179 static inline struct mlx5_ib_cq *to_mibcq(struct mlx5_core_cq *mcq) 1180 { 1181 return container_of(mcq, struct mlx5_ib_cq, mcq); 1182 } 1183 1184 static inline struct mlx5_ib_xrcd *to_mxrcd(struct ib_xrcd *ibxrcd) 1185 { 1186 return container_of(ibxrcd, struct mlx5_ib_xrcd, ibxrcd); 1187 } 1188 1189 static inline struct mlx5_ib_dev *to_mdev(struct ib_device *ibdev) 1190 { 1191 return container_of(ibdev, struct mlx5_ib_dev, ib_dev); 1192 } 1193 1194 static inline struct mlx5_ib_dev *mr_to_mdev(struct mlx5_ib_mr *mr) 1195 { 1196 return to_mdev(mr->ibmr.device); 1197 } 1198 1199 static inline struct mlx5_ib_dev *mlx5_udata_to_mdev(struct ib_udata *udata) 1200 { 1201 struct mlx5_ib_ucontext *context = rdma_udata_to_drv_context( 1202 udata, struct mlx5_ib_ucontext, ibucontext); 1203 1204 return to_mdev(context->ibucontext.device); 1205 } 1206 1207 static inline struct mlx5_ib_cq *to_mcq(struct ib_cq *ibcq) 1208 { 1209 return container_of(ibcq, struct mlx5_ib_cq, ibcq); 1210 } 1211 1212 static inline struct mlx5_ib_qp *to_mibqp(struct mlx5_core_qp *mqp) 1213 { 1214 return container_of(mqp, struct mlx5_ib_qp_base, mqp)->container_mibqp; 1215 } 1216 1217 static inline struct mlx5_ib_rwq *to_mibrwq(struct mlx5_core_qp *core_qp) 1218 { 1219 return container_of(core_qp, struct mlx5_ib_rwq, core_qp); 1220 } 1221 1222 static inline struct mlx5_ib_pd *to_mpd(struct ib_pd *ibpd) 1223 { 1224 return container_of(ibpd, struct mlx5_ib_pd, ibpd); 1225 } 1226 1227 static inline struct mlx5_ib_srq *to_msrq(struct ib_srq *ibsrq) 1228 { 1229 return container_of(ibsrq, struct mlx5_ib_srq, ibsrq); 1230 } 1231 1232 static inline struct mlx5_ib_qp *to_mqp(struct ib_qp *ibqp) 1233 { 1234 return container_of(ibqp, struct mlx5_ib_qp, ibqp); 1235 } 1236 1237 static inline struct mlx5_ib_rwq *to_mrwq(struct ib_wq *ibwq) 1238 { 1239 return container_of(ibwq, struct mlx5_ib_rwq, ibwq); 1240 } 1241 1242 static inline struct mlx5_ib_rwq_ind_table *to_mrwq_ind_table(struct ib_rwq_ind_table *ib_rwq_ind_tbl) 1243 { 1244 return container_of(ib_rwq_ind_tbl, struct mlx5_ib_rwq_ind_table, ib_rwq_ind_tbl); 1245 } 1246 1247 static inline struct mlx5_ib_srq *to_mibsrq(struct mlx5_core_srq *msrq) 1248 { 1249 return container_of(msrq, struct mlx5_ib_srq, msrq); 1250 } 1251 1252 static inline struct mlx5_ib_mr *to_mmr(struct ib_mr *ibmr) 1253 { 1254 return container_of(ibmr, struct mlx5_ib_mr, ibmr); 1255 } 1256 1257 static inline struct mlx5_ib_mw *to_mmw(struct ib_mw *ibmw) 1258 { 1259 return container_of(ibmw, struct mlx5_ib_mw, ibmw); 1260 } 1261 1262 static inline struct mlx5_ib_flow_action * 1263 to_mflow_act(struct ib_flow_action *ibact) 1264 { 1265 return container_of(ibact, struct mlx5_ib_flow_action, ib_action); 1266 } 1267 1268 static inline struct mlx5_user_mmap_entry * 1269 to_mmmap(struct rdma_user_mmap_entry *rdma_entry) 1270 { 1271 return container_of(rdma_entry, 1272 struct mlx5_user_mmap_entry, rdma_entry); 1273 } 1274 1275 int mlx5_ib_dev_res_cq_init(struct mlx5_ib_dev *dev); 1276 int mlx5_ib_dev_res_srq_init(struct mlx5_ib_dev *dev); 1277 int mlx5_ib_db_map_user(struct mlx5_ib_ucontext *context, unsigned long virt, 1278 struct mlx5_db *db); 1279 void mlx5_ib_db_unmap_user(struct mlx5_ib_ucontext *context, struct mlx5_db *db); 1280 void __mlx5_ib_cq_clean(struct mlx5_ib_cq *cq, u32 qpn, struct mlx5_ib_srq *srq); 1281 void mlx5_ib_cq_clean(struct mlx5_ib_cq *cq, u32 qpn, struct mlx5_ib_srq *srq); 1282 void mlx5_ib_free_srq_wqe(struct mlx5_ib_srq *srq, int wqe_index); 1283 int mlx5_ib_create_ah(struct ib_ah *ah, struct rdma_ah_init_attr *init_attr, 1284 struct ib_udata *udata); 1285 int mlx5_ib_query_ah(struct ib_ah *ibah, struct rdma_ah_attr *ah_attr); 1286 static inline int mlx5_ib_destroy_ah(struct ib_ah *ah, u32 flags) 1287 { 1288 return 0; 1289 } 1290 int mlx5_ib_create_srq(struct ib_srq *srq, struct ib_srq_init_attr *init_attr, 1291 struct ib_udata *udata); 1292 int mlx5_ib_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr, 1293 enum ib_srq_attr_mask attr_mask, struct ib_udata *udata); 1294 int mlx5_ib_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *srq_attr); 1295 int mlx5_ib_destroy_srq(struct ib_srq *srq, struct ib_udata *udata); 1296 int mlx5_ib_post_srq_recv(struct ib_srq *ibsrq, const struct ib_recv_wr *wr, 1297 const struct ib_recv_wr **bad_wr); 1298 int mlx5_ib_enable_lb(struct mlx5_ib_dev *dev, bool td, bool qp); 1299 void mlx5_ib_disable_lb(struct mlx5_ib_dev *dev, bool td, bool qp); 1300 int mlx5_ib_create_qp(struct ib_qp *qp, struct ib_qp_init_attr *init_attr, 1301 struct ib_udata *udata); 1302 int mlx5_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, 1303 int attr_mask, struct ib_udata *udata); 1304 int mlx5_ib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr, int qp_attr_mask, 1305 struct ib_qp_init_attr *qp_init_attr); 1306 int mlx5_ib_destroy_qp(struct ib_qp *qp, struct ib_udata *udata); 1307 void mlx5_ib_drain_sq(struct ib_qp *qp); 1308 void mlx5_ib_drain_rq(struct ib_qp *qp); 1309 int mlx5_ib_read_wqe_sq(struct mlx5_ib_qp *qp, int wqe_index, void *buffer, 1310 size_t buflen, size_t *bc); 1311 int mlx5_ib_read_wqe_rq(struct mlx5_ib_qp *qp, int wqe_index, void *buffer, 1312 size_t buflen, size_t *bc); 1313 int mlx5_ib_read_wqe_srq(struct mlx5_ib_srq *srq, int wqe_index, void *buffer, 1314 size_t buflen, size_t *bc); 1315 int mlx5_ib_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr, 1316 struct ib_udata *udata); 1317 int mlx5_ib_destroy_cq(struct ib_cq *cq, struct ib_udata *udata); 1318 int mlx5_ib_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc); 1319 int mlx5_ib_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags); 1320 int mlx5_ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period); 1321 int mlx5_ib_resize_cq(struct ib_cq *ibcq, int entries, struct ib_udata *udata); 1322 struct ib_mr *mlx5_ib_get_dma_mr(struct ib_pd *pd, int acc); 1323 struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, 1324 u64 virt_addr, int access_flags, 1325 struct ib_udata *udata); 1326 struct ib_mr *mlx5_ib_reg_user_mr_dmabuf(struct ib_pd *pd, u64 start, 1327 u64 length, u64 virt_addr, 1328 int fd, int access_flags, 1329 struct ib_udata *udata); 1330 int mlx5_ib_advise_mr(struct ib_pd *pd, 1331 enum ib_uverbs_advise_mr_advice advice, 1332 u32 flags, 1333 struct ib_sge *sg_list, 1334 u32 num_sge, 1335 struct uverbs_attr_bundle *attrs); 1336 int mlx5_ib_alloc_mw(struct ib_mw *mw, struct ib_udata *udata); 1337 int mlx5_ib_dealloc_mw(struct ib_mw *mw); 1338 struct mlx5_ib_mr *mlx5_ib_alloc_implicit_mr(struct mlx5_ib_pd *pd, 1339 int access_flags); 1340 void mlx5_ib_free_implicit_mr(struct mlx5_ib_mr *mr); 1341 void mlx5_ib_free_odp_mr(struct mlx5_ib_mr *mr); 1342 struct ib_mr *mlx5_ib_rereg_user_mr(struct ib_mr *ib_mr, int flags, u64 start, 1343 u64 length, u64 virt_addr, int access_flags, 1344 struct ib_pd *pd, struct ib_udata *udata); 1345 int mlx5_ib_dereg_mr(struct ib_mr *ibmr, struct ib_udata *udata); 1346 struct ib_mr *mlx5_ib_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type, 1347 u32 max_num_sg); 1348 struct ib_mr *mlx5_ib_alloc_mr_integrity(struct ib_pd *pd, 1349 u32 max_num_sg, 1350 u32 max_num_meta_sg); 1351 int mlx5_ib_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg, int sg_nents, 1352 unsigned int *sg_offset); 1353 int mlx5_ib_map_mr_sg_pi(struct ib_mr *ibmr, struct scatterlist *data_sg, 1354 int data_sg_nents, unsigned int *data_sg_offset, 1355 struct scatterlist *meta_sg, int meta_sg_nents, 1356 unsigned int *meta_sg_offset); 1357 int mlx5_ib_process_mad(struct ib_device *ibdev, int mad_flags, u32 port_num, 1358 const struct ib_wc *in_wc, const struct ib_grh *in_grh, 1359 const struct ib_mad *in, struct ib_mad *out, 1360 size_t *out_mad_size, u16 *out_mad_pkey_index); 1361 int mlx5_ib_alloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata); 1362 int mlx5_ib_dealloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata); 1363 int mlx5_query_ext_port_caps(struct mlx5_ib_dev *dev, unsigned int port); 1364 int mlx5_query_mad_ifc_system_image_guid(struct ib_device *ibdev, 1365 __be64 *sys_image_guid); 1366 int mlx5_query_mad_ifc_max_pkeys(struct ib_device *ibdev, 1367 u16 *max_pkeys); 1368 int mlx5_query_mad_ifc_vendor_id(struct ib_device *ibdev, 1369 u32 *vendor_id); 1370 int mlx5_query_mad_ifc_node_desc(struct mlx5_ib_dev *dev, char *node_desc); 1371 int mlx5_query_mad_ifc_node_guid(struct mlx5_ib_dev *dev, __be64 *node_guid); 1372 int mlx5_query_mad_ifc_pkey(struct ib_device *ibdev, u32 port, u16 index, 1373 u16 *pkey); 1374 int mlx5_query_mad_ifc_gids(struct ib_device *ibdev, u32 port, int index, 1375 union ib_gid *gid); 1376 int mlx5_query_mad_ifc_port(struct ib_device *ibdev, u32 port, 1377 struct ib_port_attr *props); 1378 int mlx5_ib_query_port(struct ib_device *ibdev, u32 port, 1379 struct ib_port_attr *props); 1380 void mlx5_ib_populate_pas(struct ib_umem *umem, size_t page_size, __be64 *pas, 1381 u64 access_flags); 1382 void mlx5_ib_copy_pas(u64 *old, u64 *new, int step, int num); 1383 int mlx5_ib_get_cqe_size(struct ib_cq *ibcq); 1384 int mlx5_mkey_cache_init(struct mlx5_ib_dev *dev); 1385 void mlx5_mkey_cache_cleanup(struct mlx5_ib_dev *dev); 1386 struct mlx5_cache_ent * 1387 mlx5r_cache_create_ent_locked(struct mlx5_ib_dev *dev, 1388 struct mlx5r_cache_rb_key rb_key, 1389 bool persistent_entry); 1390 1391 struct mlx5_ib_mr *mlx5_mr_cache_alloc(struct mlx5_ib_dev *dev, 1392 int access_flags, int access_mode, 1393 int ndescs); 1394 1395 int mlx5_ib_check_mr_status(struct ib_mr *ibmr, u32 check_mask, 1396 struct ib_mr_status *mr_status); 1397 struct ib_wq *mlx5_ib_create_wq(struct ib_pd *pd, 1398 struct ib_wq_init_attr *init_attr, 1399 struct ib_udata *udata); 1400 int mlx5_ib_destroy_wq(struct ib_wq *wq, struct ib_udata *udata); 1401 int mlx5_ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *wq_attr, 1402 u32 wq_attr_mask, struct ib_udata *udata); 1403 int mlx5_ib_create_rwq_ind_table(struct ib_rwq_ind_table *ib_rwq_ind_table, 1404 struct ib_rwq_ind_table_init_attr *init_attr, 1405 struct ib_udata *udata); 1406 int mlx5_ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table); 1407 struct ib_mr *mlx5_ib_reg_dm_mr(struct ib_pd *pd, struct ib_dm *dm, 1408 struct ib_dm_mr_attr *attr, 1409 struct uverbs_attr_bundle *attrs); 1410 1411 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING 1412 int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev); 1413 int mlx5r_odp_create_eq(struct mlx5_ib_dev *dev, struct mlx5_ib_pf_eq *eq); 1414 void mlx5_ib_odp_cleanup_one(struct mlx5_ib_dev *ibdev); 1415 int __init mlx5_ib_odp_init(void); 1416 void mlx5_ib_odp_cleanup(void); 1417 int mlx5_odp_init_mkey_cache(struct mlx5_ib_dev *dev); 1418 void mlx5_odp_populate_xlt(void *xlt, size_t idx, size_t nentries, 1419 struct mlx5_ib_mr *mr, int flags); 1420 1421 int mlx5_ib_advise_mr_prefetch(struct ib_pd *pd, 1422 enum ib_uverbs_advise_mr_advice advice, 1423 u32 flags, struct ib_sge *sg_list, u32 num_sge); 1424 int mlx5_ib_init_odp_mr(struct mlx5_ib_mr *mr); 1425 int mlx5_ib_init_dmabuf_mr(struct mlx5_ib_mr *mr); 1426 #else /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */ 1427 static inline int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev) { return 0; } 1428 static inline int mlx5r_odp_create_eq(struct mlx5_ib_dev *dev, 1429 struct mlx5_ib_pf_eq *eq) 1430 { 1431 return 0; 1432 } 1433 static inline void mlx5_ib_odp_cleanup_one(struct mlx5_ib_dev *ibdev) {} 1434 static inline int mlx5_ib_odp_init(void) { return 0; } 1435 static inline void mlx5_ib_odp_cleanup(void) {} 1436 static inline int mlx5_odp_init_mkey_cache(struct mlx5_ib_dev *dev) 1437 { 1438 return 0; 1439 } 1440 static inline void mlx5_odp_populate_xlt(void *xlt, size_t idx, size_t nentries, 1441 struct mlx5_ib_mr *mr, int flags) {} 1442 1443 static inline int 1444 mlx5_ib_advise_mr_prefetch(struct ib_pd *pd, 1445 enum ib_uverbs_advise_mr_advice advice, u32 flags, 1446 struct ib_sge *sg_list, u32 num_sge) 1447 { 1448 return -EOPNOTSUPP; 1449 } 1450 static inline int mlx5_ib_init_odp_mr(struct mlx5_ib_mr *mr) 1451 { 1452 return -EOPNOTSUPP; 1453 } 1454 static inline int mlx5_ib_init_dmabuf_mr(struct mlx5_ib_mr *mr) 1455 { 1456 return -EOPNOTSUPP; 1457 } 1458 #endif /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */ 1459 1460 extern const struct mmu_interval_notifier_ops mlx5_mn_ops; 1461 1462 /* Needed for rep profile */ 1463 void __mlx5_ib_remove(struct mlx5_ib_dev *dev, 1464 const struct mlx5_ib_profile *profile, 1465 int stage); 1466 int __mlx5_ib_add(struct mlx5_ib_dev *dev, 1467 const struct mlx5_ib_profile *profile); 1468 1469 int mlx5_ib_get_vf_config(struct ib_device *device, int vf, 1470 u32 port, struct ifla_vf_info *info); 1471 int mlx5_ib_set_vf_link_state(struct ib_device *device, int vf, 1472 u32 port, int state); 1473 int mlx5_ib_get_vf_stats(struct ib_device *device, int vf, 1474 u32 port, struct ifla_vf_stats *stats); 1475 int mlx5_ib_get_vf_guid(struct ib_device *device, int vf, u32 port, 1476 struct ifla_vf_guid *node_guid, 1477 struct ifla_vf_guid *port_guid); 1478 int mlx5_ib_set_vf_guid(struct ib_device *device, int vf, u32 port, 1479 u64 guid, int type); 1480 1481 __be16 mlx5_get_roce_udp_sport_min(const struct mlx5_ib_dev *dev, 1482 const struct ib_gid_attr *attr); 1483 1484 void mlx5_ib_cleanup_cong_debugfs(struct mlx5_ib_dev *dev, u32 port_num); 1485 void mlx5_ib_init_cong_debugfs(struct mlx5_ib_dev *dev, u32 port_num); 1486 1487 /* GSI QP helper functions */ 1488 int mlx5_ib_create_gsi(struct ib_pd *pd, struct mlx5_ib_qp *mqp, 1489 struct ib_qp_init_attr *attr); 1490 int mlx5_ib_destroy_gsi(struct mlx5_ib_qp *mqp); 1491 int mlx5_ib_gsi_modify_qp(struct ib_qp *qp, struct ib_qp_attr *attr, 1492 int attr_mask); 1493 int mlx5_ib_gsi_query_qp(struct ib_qp *qp, struct ib_qp_attr *qp_attr, 1494 int qp_attr_mask, 1495 struct ib_qp_init_attr *qp_init_attr); 1496 int mlx5_ib_gsi_post_send(struct ib_qp *qp, const struct ib_send_wr *wr, 1497 const struct ib_send_wr **bad_wr); 1498 int mlx5_ib_gsi_post_recv(struct ib_qp *qp, const struct ib_recv_wr *wr, 1499 const struct ib_recv_wr **bad_wr); 1500 void mlx5_ib_gsi_pkey_change(struct mlx5_ib_gsi_qp *gsi); 1501 1502 int mlx5_ib_generate_wc(struct ib_cq *ibcq, struct ib_wc *wc); 1503 1504 void mlx5_ib_free_bfreg(struct mlx5_ib_dev *dev, struct mlx5_bfreg_info *bfregi, 1505 int bfregn); 1506 struct mlx5_ib_dev *mlx5_ib_get_ibdev_from_mpi(struct mlx5_ib_multiport_info *mpi); 1507 struct mlx5_core_dev *mlx5_ib_get_native_port_mdev(struct mlx5_ib_dev *dev, 1508 u32 ib_port_num, 1509 u32 *native_port_num); 1510 void mlx5_ib_put_native_port_mdev(struct mlx5_ib_dev *dev, 1511 u32 port_num); 1512 1513 extern const struct uapi_definition mlx5_ib_devx_defs[]; 1514 extern const struct uapi_definition mlx5_ib_flow_defs[]; 1515 extern const struct uapi_definition mlx5_ib_qos_defs[]; 1516 extern const struct uapi_definition mlx5_ib_std_types_defs[]; 1517 1518 static inline int is_qp1(enum ib_qp_type qp_type) 1519 { 1520 return qp_type == MLX5_IB_QPT_HW_GSI || qp_type == IB_QPT_GSI; 1521 } 1522 1523 static inline u32 check_cq_create_flags(u32 flags) 1524 { 1525 /* 1526 * It returns non-zero value for unsupported CQ 1527 * create flags, otherwise it returns zero. 1528 */ 1529 return (flags & ~(IB_UVERBS_CQ_FLAGS_IGNORE_OVERRUN | 1530 IB_UVERBS_CQ_FLAGS_TIMESTAMP_COMPLETION)); 1531 } 1532 1533 static inline int verify_assign_uidx(u8 cqe_version, u32 cmd_uidx, 1534 u32 *user_index) 1535 { 1536 if (cqe_version) { 1537 if ((cmd_uidx == MLX5_IB_DEFAULT_UIDX) || 1538 (cmd_uidx & ~MLX5_USER_ASSIGNED_UIDX_MASK)) 1539 return -EINVAL; 1540 *user_index = cmd_uidx; 1541 } else { 1542 *user_index = MLX5_IB_DEFAULT_UIDX; 1543 } 1544 1545 return 0; 1546 } 1547 1548 static inline int get_qp_user_index(struct mlx5_ib_ucontext *ucontext, 1549 struct mlx5_ib_create_qp *ucmd, 1550 int inlen, 1551 u32 *user_index) 1552 { 1553 u8 cqe_version = ucontext->cqe_version; 1554 1555 if ((offsetofend(typeof(*ucmd), uidx) <= inlen) && !cqe_version && 1556 (ucmd->uidx == MLX5_IB_DEFAULT_UIDX)) 1557 return 0; 1558 1559 if ((offsetofend(typeof(*ucmd), uidx) <= inlen) != !!cqe_version) 1560 return -EINVAL; 1561 1562 return verify_assign_uidx(cqe_version, ucmd->uidx, user_index); 1563 } 1564 1565 static inline int get_srq_user_index(struct mlx5_ib_ucontext *ucontext, 1566 struct mlx5_ib_create_srq *ucmd, 1567 int inlen, 1568 u32 *user_index) 1569 { 1570 u8 cqe_version = ucontext->cqe_version; 1571 1572 if ((offsetofend(typeof(*ucmd), uidx) <= inlen) && !cqe_version && 1573 (ucmd->uidx == MLX5_IB_DEFAULT_UIDX)) 1574 return 0; 1575 1576 if ((offsetofend(typeof(*ucmd), uidx) <= inlen) != !!cqe_version) 1577 return -EINVAL; 1578 1579 return verify_assign_uidx(cqe_version, ucmd->uidx, user_index); 1580 } 1581 1582 static inline int get_uars_per_sys_page(struct mlx5_ib_dev *dev, bool lib_support) 1583 { 1584 return lib_support && MLX5_CAP_GEN(dev->mdev, uar_4k) ? 1585 MLX5_UARS_IN_PAGE : 1; 1586 } 1587 1588 extern void *xlt_emergency_page; 1589 1590 int bfregn_to_uar_index(struct mlx5_ib_dev *dev, 1591 struct mlx5_bfreg_info *bfregi, u32 bfregn, 1592 bool dyn_bfreg); 1593 1594 static inline int mlx5r_store_odp_mkey(struct mlx5_ib_dev *dev, 1595 struct mlx5_ib_mkey *mmkey) 1596 { 1597 refcount_set(&mmkey->usecount, 1); 1598 1599 return xa_err(xa_store(&dev->odp_mkeys, mlx5_base_mkey(mmkey->key), 1600 mmkey, GFP_KERNEL)); 1601 } 1602 1603 /* deref an mkey that can participate in ODP flow */ 1604 static inline void mlx5r_deref_odp_mkey(struct mlx5_ib_mkey *mmkey) 1605 { 1606 if (refcount_dec_and_test(&mmkey->usecount)) 1607 wake_up(&mmkey->wait); 1608 } 1609 1610 /* deref an mkey that can participate in ODP flow and wait for relese */ 1611 static inline void mlx5r_deref_wait_odp_mkey(struct mlx5_ib_mkey *mmkey) 1612 { 1613 mlx5r_deref_odp_mkey(mmkey); 1614 wait_event(mmkey->wait, refcount_read(&mmkey->usecount) == 0); 1615 } 1616 1617 int mlx5_ib_test_wc(struct mlx5_ib_dev *dev); 1618 1619 static inline bool mlx5_ib_lag_should_assign_affinity(struct mlx5_ib_dev *dev) 1620 { 1621 /* 1622 * If the driver is in hash mode and the port_select_flow_table_bypass cap 1623 * is supported, it means that the driver no longer needs to assign the port 1624 * affinity by default. If a user wants to set the port affinity explicitly, 1625 * the user has a dedicated API to do that, so there is no need to assign 1626 * the port affinity by default. 1627 */ 1628 if (dev->lag_active && 1629 mlx5_lag_mode_is_hash(dev->mdev) && 1630 MLX5_CAP_PORT_SELECTION(dev->mdev, port_select_flow_table_bypass)) 1631 return 0; 1632 1633 if (mlx5_lag_is_lacp_owner(dev->mdev) && !dev->lag_active) 1634 return 0; 1635 1636 return dev->lag_active || 1637 (MLX5_CAP_GEN(dev->mdev, num_lag_ports) > 1 && 1638 MLX5_CAP_GEN(dev->mdev, lag_tx_port_affinity)); 1639 } 1640 1641 static inline bool rt_supported(int ts_cap) 1642 { 1643 return ts_cap == MLX5_TIMESTAMP_FORMAT_CAP_REAL_TIME || 1644 ts_cap == MLX5_TIMESTAMP_FORMAT_CAP_FREE_RUNNING_AND_REAL_TIME; 1645 } 1646 1647 /* 1648 * PCI Peer to Peer is a trainwreck. If no switch is present then things 1649 * sometimes work, depending on the pci_distance_p2p logic for excluding broken 1650 * root complexes. However if a switch is present in the path, then things get 1651 * really ugly depending on how the switch is setup. This table assumes that the 1652 * root complex is strict and is validating that all req/reps are matches 1653 * perfectly - so any scenario where it sees only half the transaction is a 1654 * failure. 1655 * 1656 * CR/RR/DT ATS RO P2P 1657 * 00X X X OK 1658 * 010 X X fails (request is routed to root but root never sees comp) 1659 * 011 0 X fails (request is routed to root but root never sees comp) 1660 * 011 1 X OK 1661 * 10X X 1 OK 1662 * 101 X 0 fails (completion is routed to root but root didn't see req) 1663 * 110 X 0 SLOW 1664 * 111 0 0 SLOW 1665 * 111 1 0 fails (completion is routed to root but root didn't see req) 1666 * 111 1 1 OK 1667 * 1668 * Unfortunately we cannot reliably know if a switch is present or what the 1669 * CR/RR/DT ACS settings are, as in a VM that is all hidden. Assume that 1670 * CR/RR/DT is 111 if the ATS cap is enabled and follow the last three rows. 1671 * 1672 * For now assume if the umem is a dma_buf then it is P2P. 1673 */ 1674 static inline bool mlx5_umem_needs_ats(struct mlx5_ib_dev *dev, 1675 struct ib_umem *umem, int access_flags) 1676 { 1677 if (!MLX5_CAP_GEN(dev->mdev, ats) || !umem->is_dmabuf) 1678 return false; 1679 return access_flags & IB_ACCESS_RELAXED_ORDERING; 1680 } 1681 1682 int set_roce_addr(struct mlx5_ib_dev *dev, u32 port_num, 1683 unsigned int index, const union ib_gid *gid, 1684 const struct ib_gid_attr *attr); 1685 #endif /* MLX5_IB_H */ 1686