1 /*- 2 * Copyright (c) 2013-2015, Mellanox Technologies, Ltd. All rights reserved. 3 * 4 * Redistribution and use in source and binary forms, with or without 5 * modification, are permitted provided that the following conditions 6 * are met: 7 * 1. Redistributions of source code must retain the above copyright 8 * notice, this list of conditions and the following disclaimer. 9 * 2. Redistributions in binary form must reproduce the above copyright 10 * notice, this list of conditions and the following disclaimer in the 11 * documentation and/or other materials provided with the distribution. 12 * 13 * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS `AS IS' AND 14 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 15 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 16 * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE 17 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 18 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 19 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 20 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 21 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 22 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 23 * SUCH DAMAGE. 24 * 25 * $FreeBSD$ 26 */ 27 28 #ifndef MLX5_IB_H 29 #define MLX5_IB_H 30 31 #include <linux/kernel.h> 32 #include <linux/sched.h> 33 #include <linux/printk.h> 34 #include <rdma/ib_verbs.h> 35 #include <rdma/ib_smi.h> 36 #include <dev/mlx5/cq.h> 37 #include <dev/mlx5/qp.h> 38 #include <dev/mlx5/srq.h> 39 #include <linux/types.h> 40 #include <dev/mlx5/mlx5_core/transobj.h> 41 #include <rdma/ib_user_verbs.h> 42 #include <rdma/mlx5-abi.h> 43 44 #define mlx5_ib_dbg(dev, format, arg...) \ 45 pr_debug("%s:%s:%d:(pid %d): " format, (dev)->ib_dev.name, __func__, \ 46 __LINE__, current->pid, ##arg) 47 48 #define mlx5_ib_err(dev, format, arg...) \ 49 pr_err("%s:%s:%d:(pid %d): " format, (dev)->ib_dev.name, __func__, \ 50 __LINE__, current->pid, ##arg) 51 52 #define mlx5_ib_warn(dev, format, arg...) \ 53 pr_warn("%s:%s:%d:(pid %d): " format, (dev)->ib_dev.name, __func__, \ 54 __LINE__, current->pid, ##arg) 55 56 #define field_avail(type, fld, sz) (offsetof(type, fld) + \ 57 sizeof(((type *)0)->fld) <= (sz)) 58 #define MLX5_IB_DEFAULT_UIDX 0xffffff 59 #define MLX5_USER_ASSIGNED_UIDX_MASK __mlx5_mask(qpc, user_index) 60 61 enum { 62 MLX5_IB_MMAP_CMD_SHIFT = 8, 63 MLX5_IB_MMAP_CMD_MASK = 0xff, 64 }; 65 66 enum mlx5_ib_mmap_cmd { 67 MLX5_IB_MMAP_REGULAR_PAGE = 0, 68 MLX5_IB_MMAP_GET_CONTIGUOUS_PAGES = 1, 69 MLX5_IB_MMAP_WC_PAGE = 2, 70 MLX5_IB_MMAP_NC_PAGE = 3, 71 /* 5 is chosen in order to be compatible with old versions of libmlx5 */ 72 MLX5_IB_MMAP_CORE_CLOCK = 5, 73 }; 74 75 enum { 76 MLX5_RES_SCAT_DATA32_CQE = 0x1, 77 MLX5_RES_SCAT_DATA64_CQE = 0x2, 78 MLX5_REQ_SCAT_DATA32_CQE = 0x11, 79 MLX5_REQ_SCAT_DATA64_CQE = 0x22, 80 }; 81 82 enum mlx5_ib_latency_class { 83 MLX5_IB_LATENCY_CLASS_LOW, 84 MLX5_IB_LATENCY_CLASS_MEDIUM, 85 MLX5_IB_LATENCY_CLASS_HIGH, 86 MLX5_IB_LATENCY_CLASS_FAST_PATH 87 }; 88 89 enum mlx5_ib_mad_ifc_flags { 90 MLX5_MAD_IFC_IGNORE_MKEY = 1, 91 MLX5_MAD_IFC_IGNORE_BKEY = 2, 92 MLX5_MAD_IFC_NET_VIEW = 4, 93 }; 94 95 enum { 96 MLX5_CROSS_CHANNEL_UUAR = 0, 97 }; 98 99 enum { 100 MLX5_CQE_VERSION_V0, 101 MLX5_CQE_VERSION_V1, 102 }; 103 104 struct mlx5_ib_vma_private_data { 105 struct list_head list; 106 struct vm_area_struct *vma; 107 }; 108 109 struct mlx5_ib_ucontext { 110 struct ib_ucontext ibucontext; 111 struct list_head db_page_list; 112 113 /* protect doorbell record alloc/free 114 */ 115 struct mutex db_page_mutex; 116 struct mlx5_uuar_info uuari; 117 u8 cqe_version; 118 /* Transport Domain number */ 119 u32 tdn; 120 struct list_head vma_private_list; 121 }; 122 123 static inline struct mlx5_ib_ucontext *to_mucontext(struct ib_ucontext *ibucontext) 124 { 125 return container_of(ibucontext, struct mlx5_ib_ucontext, ibucontext); 126 } 127 128 struct mlx5_ib_pd { 129 struct ib_pd ibpd; 130 u32 pdn; 131 }; 132 133 #define MLX5_IB_FLOW_MCAST_PRIO (MLX5_BY_PASS_NUM_PRIOS - 1) 134 #define MLX5_IB_FLOW_LAST_PRIO (MLX5_BY_PASS_NUM_REGULAR_PRIOS - 1) 135 #if (MLX5_IB_FLOW_LAST_PRIO <= 0) 136 #error "Invalid number of bypass priorities" 137 #endif 138 #define MLX5_IB_FLOW_LEFTOVERS_PRIO (MLX5_IB_FLOW_MCAST_PRIO + 1) 139 140 #define MLX5_IB_NUM_FLOW_FT (MLX5_IB_FLOW_LEFTOVERS_PRIO + 1) 141 #define MLX5_IB_NUM_SNIFFER_FTS 2 142 struct mlx5_ib_flow_prio { 143 struct mlx5_flow_table *flow_table; 144 unsigned int refcount; 145 }; 146 147 struct mlx5_ib_flow_handler { 148 struct list_head list; 149 struct ib_flow ibflow; 150 struct mlx5_ib_flow_prio *prio; 151 struct mlx5_flow_rule *rule; 152 }; 153 154 struct mlx5_ib_flow_db { 155 struct mlx5_ib_flow_prio prios[MLX5_IB_NUM_FLOW_FT]; 156 struct mlx5_ib_flow_prio sniffer[MLX5_IB_NUM_SNIFFER_FTS]; 157 struct mlx5_flow_table *lag_demux_ft; 158 /* Protect flow steering bypass flow tables 159 * when add/del flow rules. 160 * only single add/removal of flow steering rule could be done 161 * simultaneously. 162 */ 163 struct mutex lock; 164 }; 165 166 /* Use macros here so that don't have to duplicate 167 * enum ib_send_flags and enum ib_qp_type for low-level driver 168 */ 169 170 #define MLX5_IB_SEND_UMR_UNREG IB_SEND_RESERVED_START 171 #define MLX5_IB_SEND_UMR_FAIL_IF_FREE (IB_SEND_RESERVED_START << 1) 172 #define MLX5_IB_SEND_UMR_UPDATE_MTT (IB_SEND_RESERVED_START << 2) 173 174 #define MLX5_IB_SEND_UMR_UPDATE_TRANSLATION (IB_SEND_RESERVED_START << 3) 175 #define MLX5_IB_SEND_UMR_UPDATE_PD (IB_SEND_RESERVED_START << 4) 176 #define MLX5_IB_SEND_UMR_UPDATE_ACCESS IB_SEND_RESERVED_END 177 178 #define MLX5_IB_QPT_REG_UMR IB_QPT_RESERVED1 179 /* 180 * IB_QPT_GSI creates the software wrapper around GSI, and MLX5_IB_QPT_HW_GSI 181 * creates the actual hardware QP. 182 */ 183 #define MLX5_IB_QPT_HW_GSI IB_QPT_RESERVED2 184 #define MLX5_IB_WR_UMR IB_WR_RESERVED1 185 186 /* Private QP creation flags to be passed in ib_qp_init_attr.create_flags. 187 * 188 * These flags are intended for internal use by the mlx5_ib driver, and they 189 * rely on the range reserved for that use in the ib_qp_create_flags enum. 190 */ 191 192 /* Create a UD QP whose source QP number is 1 */ 193 static inline enum ib_qp_create_flags mlx5_ib_create_qp_sqpn_qp1(void) 194 { 195 return IB_QP_CREATE_RESERVED_START; 196 } 197 198 struct wr_list { 199 u16 opcode; 200 u16 next; 201 }; 202 203 struct mlx5_ib_wq { 204 u64 *wrid; 205 u32 *wr_data; 206 struct wr_list *w_list; 207 unsigned *wqe_head; 208 u16 unsig_count; 209 210 /* serialize post to the work queue 211 */ 212 spinlock_t lock; 213 int wqe_cnt; 214 int max_post; 215 int max_gs; 216 int offset; 217 int wqe_shift; 218 unsigned head; 219 unsigned tail; 220 u16 cur_post; 221 u16 last_poll; 222 void *qend; 223 }; 224 225 struct mlx5_ib_rwq { 226 struct ib_wq ibwq; 227 struct mlx5_core_qp core_qp; 228 u32 rq_num_pas; 229 u32 log_rq_stride; 230 u32 log_rq_size; 231 u32 rq_page_offset; 232 u32 log_page_size; 233 struct ib_umem *umem; 234 size_t buf_size; 235 unsigned int page_shift; 236 int create_type; 237 struct mlx5_db db; 238 u32 user_index; 239 u32 wqe_count; 240 u32 wqe_shift; 241 int wq_sig; 242 }; 243 244 enum { 245 MLX5_QP_USER, 246 MLX5_QP_KERNEL, 247 MLX5_QP_EMPTY 248 }; 249 250 enum { 251 MLX5_WQ_USER, 252 MLX5_WQ_KERNEL 253 }; 254 255 struct mlx5_ib_rwq_ind_table { 256 struct ib_rwq_ind_table ib_rwq_ind_tbl; 257 u32 rqtn; 258 }; 259 260 /* 261 * Connect-IB can trigger up to four concurrent pagefaults 262 * per-QP. 263 */ 264 enum mlx5_ib_pagefault_context { 265 MLX5_IB_PAGEFAULT_RESPONDER_READ, 266 MLX5_IB_PAGEFAULT_REQUESTOR_READ, 267 MLX5_IB_PAGEFAULT_RESPONDER_WRITE, 268 MLX5_IB_PAGEFAULT_REQUESTOR_WRITE, 269 MLX5_IB_PAGEFAULT_CONTEXTS 270 }; 271 272 static inline enum mlx5_ib_pagefault_context 273 mlx5_ib_get_pagefault_context(struct mlx5_pagefault *pagefault) 274 { 275 return pagefault->flags & (MLX5_PFAULT_REQUESTOR | MLX5_PFAULT_WRITE); 276 } 277 278 struct mlx5_ib_pfault { 279 struct work_struct work; 280 struct mlx5_pagefault mpfault; 281 }; 282 283 struct mlx5_ib_ubuffer { 284 struct ib_umem *umem; 285 int buf_size; 286 u64 buf_addr; 287 }; 288 289 struct mlx5_ib_qp_base { 290 struct mlx5_ib_qp *container_mibqp; 291 struct mlx5_core_qp mqp; 292 struct mlx5_ib_ubuffer ubuffer; 293 }; 294 295 struct mlx5_ib_qp_trans { 296 struct mlx5_ib_qp_base base; 297 u16 xrcdn; 298 u8 alt_port; 299 u8 atomic_rd_en; 300 u8 resp_depth; 301 }; 302 303 struct mlx5_ib_rss_qp { 304 u32 tirn; 305 }; 306 307 struct mlx5_ib_rq { 308 struct mlx5_ib_qp_base base; 309 struct mlx5_ib_wq *rq; 310 struct mlx5_ib_ubuffer ubuffer; 311 struct mlx5_db *doorbell; 312 u32 tirn; 313 u8 state; 314 }; 315 316 struct mlx5_ib_sq { 317 struct mlx5_ib_qp_base base; 318 struct mlx5_ib_wq *sq; 319 struct mlx5_ib_ubuffer ubuffer; 320 struct mlx5_db *doorbell; 321 u32 tisn; 322 u8 state; 323 }; 324 325 struct mlx5_ib_raw_packet_qp { 326 struct mlx5_ib_sq sq; 327 struct mlx5_ib_rq rq; 328 }; 329 330 struct mlx5_ib_qp { 331 struct ib_qp ibqp; 332 union { 333 struct mlx5_ib_qp_trans trans_qp; 334 struct mlx5_ib_raw_packet_qp raw_packet_qp; 335 struct mlx5_ib_rss_qp rss_qp; 336 }; 337 struct mlx5_buf buf; 338 339 struct mlx5_db db; 340 struct mlx5_ib_wq rq; 341 342 u8 sq_signal_bits; 343 u8 fm_cache; 344 struct mlx5_ib_wq sq; 345 346 /* serialize qp state modifications 347 */ 348 struct mutex mutex; 349 u32 flags; 350 u8 port; 351 u8 state; 352 int wq_sig; 353 int scat_cqe; 354 int max_inline_data; 355 struct mlx5_bf *bf; 356 int has_rq; 357 358 /* only for user space QPs. For kernel 359 * we have it from the bf object 360 */ 361 int uuarn; 362 363 int create_type; 364 365 /* Store signature errors */ 366 bool signature_en; 367 368 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING 369 /* 370 * A flag that is true for QP's that are in a state that doesn't 371 * allow page faults, and shouldn't schedule any more faults. 372 */ 373 int disable_page_faults; 374 /* 375 * The disable_page_faults_lock protects a QP's disable_page_faults 376 * field, allowing for a thread to atomically check whether the QP 377 * allows page faults, and if so schedule a page fault. 378 */ 379 spinlock_t disable_page_faults_lock; 380 struct mlx5_ib_pfault pagefaults[MLX5_IB_PAGEFAULT_CONTEXTS]; 381 #endif 382 struct list_head qps_list; 383 struct list_head cq_recv_list; 384 struct list_head cq_send_list; 385 }; 386 387 struct mlx5_ib_cq_buf { 388 struct mlx5_buf buf; 389 struct ib_umem *umem; 390 int cqe_size; 391 int nent; 392 }; 393 394 enum mlx5_ib_qp_flags { 395 MLX5_IB_QP_LSO = IB_QP_CREATE_IPOIB_UD_LSO, 396 MLX5_IB_QP_BLOCK_MULTICAST_LOOPBACK = IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK, 397 MLX5_IB_QP_CROSS_CHANNEL = IB_QP_CREATE_CROSS_CHANNEL, 398 MLX5_IB_QP_MANAGED_SEND = IB_QP_CREATE_MANAGED_SEND, 399 MLX5_IB_QP_MANAGED_RECV = IB_QP_CREATE_MANAGED_RECV, 400 MLX5_IB_QP_SIGNATURE_HANDLING = 1 << 5, 401 /* QP uses 1 as its source QP number */ 402 MLX5_IB_QP_SQPN_QP1 = 1 << 6, 403 MLX5_IB_QP_CAP_SCATTER_FCS = 1 << 7, 404 MLX5_IB_QP_RSS = 1 << 8, 405 }; 406 407 struct mlx5_umr_wr { 408 struct ib_send_wr wr; 409 union { 410 u64 virt_addr; 411 u64 offset; 412 } target; 413 struct ib_pd *pd; 414 unsigned int page_shift; 415 unsigned int npages; 416 u32 length; 417 int access_flags; 418 u32 mkey; 419 }; 420 421 static inline struct mlx5_umr_wr *umr_wr(struct ib_send_wr *wr) 422 { 423 return container_of(wr, struct mlx5_umr_wr, wr); 424 } 425 426 struct mlx5_shared_mr_info { 427 int mr_id; 428 struct ib_umem *umem; 429 }; 430 431 struct mlx5_ib_cq { 432 struct ib_cq ibcq; 433 struct mlx5_core_cq mcq; 434 struct mlx5_ib_cq_buf buf; 435 struct mlx5_db db; 436 437 /* serialize access to the CQ 438 */ 439 spinlock_t lock; 440 441 /* protect resize cq 442 */ 443 struct mutex resize_mutex; 444 struct mlx5_ib_cq_buf *resize_buf; 445 struct ib_umem *resize_umem; 446 int cqe_size; 447 struct list_head list_send_qp; 448 struct list_head list_recv_qp; 449 u32 create_flags; 450 struct list_head wc_list; 451 enum ib_cq_notify_flags notify_flags; 452 struct work_struct notify_work; 453 }; 454 455 struct mlx5_ib_wc { 456 struct ib_wc wc; 457 struct list_head list; 458 }; 459 460 struct mlx5_ib_srq { 461 struct ib_srq ibsrq; 462 struct mlx5_core_srq msrq; 463 struct mlx5_buf buf; 464 struct mlx5_db db; 465 u64 *wrid; 466 /* protect SRQ hanlding 467 */ 468 spinlock_t lock; 469 int head; 470 int tail; 471 u16 wqe_ctr; 472 struct ib_umem *umem; 473 /* serialize arming a SRQ 474 */ 475 struct mutex mutex; 476 int wq_sig; 477 }; 478 479 struct mlx5_ib_xrcd { 480 struct ib_xrcd ibxrcd; 481 u32 xrcdn; 482 }; 483 484 enum mlx5_ib_mtt_access_flags { 485 MLX5_IB_MTT_READ = (1 << 0), 486 MLX5_IB_MTT_WRITE = (1 << 1), 487 }; 488 489 #define MLX5_IB_MTT_PRESENT (MLX5_IB_MTT_READ | MLX5_IB_MTT_WRITE) 490 491 struct mlx5_ib_mr { 492 struct ib_mr ibmr; 493 void *descs; 494 dma_addr_t desc_map; 495 int ndescs; 496 int max_descs; 497 int desc_size; 498 int access_mode; 499 struct mlx5_core_mr mmkey; 500 struct ib_umem *umem; 501 struct mlx5_shared_mr_info *smr_info; 502 struct list_head list; 503 int order; 504 int umred; 505 int npages; 506 struct mlx5_ib_dev *dev; 507 u32 out[MLX5_ST_SZ_DW(create_mkey_out)]; 508 struct mlx5_core_sig_ctx *sig; 509 int live; 510 void *descs_alloc; 511 int access_flags; /* Needed for rereg MR */ 512 }; 513 514 struct mlx5_ib_mw { 515 struct ib_mw ibmw; 516 struct mlx5_core_mr mmkey; 517 }; 518 519 struct mlx5_ib_umr_context { 520 struct ib_cqe cqe; 521 enum ib_wc_status status; 522 struct completion done; 523 }; 524 525 struct umr_common { 526 struct ib_pd *pd; 527 struct ib_cq *cq; 528 struct ib_qp *qp; 529 /* control access to UMR QP 530 */ 531 struct semaphore sem; 532 }; 533 534 enum { 535 MLX5_FMR_INVALID, 536 MLX5_FMR_VALID, 537 MLX5_FMR_BUSY, 538 }; 539 540 struct mlx5_cache_ent { 541 struct list_head head; 542 /* sync access to the cahce entry 543 */ 544 spinlock_t lock; 545 546 547 struct dentry *dir; 548 char name[4]; 549 u32 order; 550 u32 size; 551 u32 cur; 552 u32 miss; 553 u32 limit; 554 555 struct dentry *fsize; 556 struct dentry *fcur; 557 struct dentry *fmiss; 558 struct dentry *flimit; 559 560 struct mlx5_ib_dev *dev; 561 struct work_struct work; 562 struct delayed_work dwork; 563 int pending; 564 }; 565 566 struct mlx5_mr_cache { 567 struct workqueue_struct *wq; 568 struct mlx5_cache_ent ent[MAX_MR_CACHE_ENTRIES]; 569 int stopped; 570 struct dentry *root; 571 unsigned long last_add; 572 }; 573 574 struct mlx5_ib_gsi_qp; 575 576 struct mlx5_ib_port_resources { 577 struct mlx5_ib_resources *devr; 578 struct mlx5_ib_gsi_qp *gsi; 579 struct work_struct pkey_change_work; 580 }; 581 582 struct mlx5_ib_resources { 583 struct ib_cq *c0; 584 struct ib_xrcd *x0; 585 struct ib_xrcd *x1; 586 struct ib_pd *p0; 587 struct ib_srq *s0; 588 struct ib_srq *s1; 589 struct mlx5_ib_port_resources ports[2]; 590 /* Protects changes to the port resources */ 591 struct mutex mutex; 592 }; 593 594 struct mlx5_ib_port { 595 u16 q_cnt_id; 596 }; 597 598 struct mlx5_roce { 599 /* Protect mlx5_ib_get_netdev from invoking dev_hold() with a NULL 600 * netdev pointer 601 */ 602 rwlock_t netdev_lock; 603 struct net_device *netdev; 604 struct notifier_block nb; 605 atomic_t next_port; 606 }; 607 608 #define MLX5_IB_STATS_COUNT(a,b,c,d) a 609 #define MLX5_IB_STATS_VAR(a,b,c,d) b; 610 #define MLX5_IB_STATS_DESC(a,b,c,d) c, d, 611 612 #define MLX5_IB_CONG_PARAMS(m) \ 613 /* ECN RP */ \ 614 m(+1, u64 rp_clamp_tgt_rate, "rp_clamp_tgt_rate", "If set, whenever a CNP is processed, the target rate is updated to be the current rate") \ 615 m(+1, u64 rp_clamp_tgt_rate_ati, "rp_clamp_tgt_rate_ati", "If set, when receiving a CNP, the target rate should be updated if the transission rate was increased due to the timer, and not only due to the byte counter") \ 616 m(+1, u64 rp_time_reset, "rp_time_reset", "Time in microseconds between rate increases if no CNPs are received") \ 617 m(+1, u64 rp_byte_reset, "rp_byte_reset", "Transmitted data in bytes between rate increases if no CNP's are received. A value of zero means disabled.") \ 618 m(+1, u64 rp_threshold, "rp_threshold", "The number of times rpByteStage or rpTimeStage can count before the RP rate control state machine advances states") \ 619 m(+1, u64 rp_ai_rate, "rp_ai_rate", "The rate, in Mbits per second, used to increase rpTargetRate in the active increase state") \ 620 m(+1, u64 rp_hai_rate, "rp_hai_rate", "The rate, in Mbits per second, used to increase rpTargetRate in the hyper increase state") \ 621 m(+1, u64 rp_min_dec_fac, "rp_min_dec_fac", "The minimum factor by which the current transmit rate can be changed when processing a CNP. Value is given as a percentage, [1 .. 100]") \ 622 m(+1, u64 rp_min_rate, "rp_min_rate", "The minimum value, in Mbps per second, for rate to limit") \ 623 m(+1, u64 rp_rate_to_set_on_first_cnp, "rp_rate_to_set_on_first_cnp", "The rate that is set for the flow when a rate limiter is allocated to it upon first CNP received, in Mbps. A value of zero means use full port speed") \ 624 m(+1, u64 rp_dce_tcp_g, "rp_dce_tcp_g", "Used to update the congestion estimator, alpha, once every dce_tcp_rtt once every dce_tcp_rtt microseconds") \ 625 m(+1, u64 rp_dce_tcp_rtt, "rp_dce_tcp_rtt", "The time between updates of the aolpha value, in microseconds") \ 626 m(+1, u64 rp_rate_reduce_monitor_period, "rp_rate_reduce_monitor_period", "The minimum time between two consecutive rate reductions for a single flow") \ 627 m(+1, u64 rp_initial_alpha_value, "rp_initial_alpha_value", "The initial value of alpha to use when receiving the first CNP for a flow") \ 628 m(+1, u64 rp_gd, "rp_gd", "If a CNP is received, the flow rate is reduced at the beginning of the next rate_reduce_monitor_period interval") \ 629 /* ECN NP */ \ 630 m(+1, u64 np_cnp_dscp, "np_cnp_dscp", "The DiffServ Code Point of the generated CNP for this port") \ 631 m(+1, u64 np_cnp_prio_mode, "np_cnp_prio_mode", "The 802.1p priority value of the generated CNP for this port") \ 632 m(+1, u64 np_cnp_prio, "np_cnp_prio", "The 802.1p priority value of the generated CNP for this port") 633 634 #define MLX5_IB_CONG_PARAMS_NUM (0 MLX5_IB_CONG_PARAMS(MLX5_IB_STATS_COUNT)) 635 636 #define MLX5_IB_CONG_STATS(m) \ 637 m(+1, u64 syndrome, "syndrome", "Syndrome number") \ 638 m(+1, u64 rp_cur_flows, "rp_cur_flows", "Number of flows limited") \ 639 m(+1, u64 sum_flows, "sum_flows", "Sum of the number of flows limited over time") \ 640 m(+1, u64 rp_cnp_ignored, "rp_cnp_ignored", "Number of CNPs and CNMs ignored") \ 641 m(+1, u64 rp_cnp_handled, "rp_cnp_handled", "Number of CNPs and CNMs successfully handled") \ 642 m(+1, u64 time_stamp, "time_stamp", "Time stamp in microseconds") \ 643 m(+1, u64 accumulators_period, "accumulators_period", "The value of X variable for accumulating counters") \ 644 m(+1, u64 np_ecn_marked_roce_packets, "np_ecn_marked_roce_packets", "Number of ECN marked packets seen") \ 645 m(+1, u64 np_cnp_sent, "np_cnp_sent", "Number of CNPs sent") 646 647 #define MLX5_IB_CONG_STATS_NUM (0 MLX5_IB_CONG_STATS(MLX5_IB_STATS_COUNT)) 648 649 struct mlx5_ib_congestion { 650 struct sysctl_ctx_list ctx; 651 struct sx lock; 652 struct delayed_work dwork; 653 u64 arg [0]; 654 MLX5_IB_CONG_PARAMS(MLX5_IB_STATS_VAR) 655 MLX5_IB_CONG_STATS(MLX5_IB_STATS_VAR) 656 }; 657 658 struct mlx5_ib_dev { 659 struct ib_device ib_dev; 660 struct mlx5_core_dev *mdev; 661 struct mlx5_roce roce; 662 MLX5_DECLARE_DOORBELL_LOCK(uar_lock); 663 int num_ports; 664 /* serialize update of capability mask 665 */ 666 struct mutex cap_mask_mutex; 667 bool ib_active; 668 struct umr_common umrc; 669 /* sync used page count stats 670 */ 671 struct mlx5_ib_resources devr; 672 struct mlx5_mr_cache cache; 673 struct timer_list delay_timer; 674 /* Prevents soft lock on massive reg MRs */ 675 struct mutex slow_path_mutex; 676 int fill_delay; 677 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING 678 struct ib_odp_caps odp_caps; 679 /* 680 * Sleepable RCU that prevents destruction of MRs while they are still 681 * being used by a page fault handler. 682 */ 683 struct srcu_struct mr_srcu; 684 #endif 685 struct mlx5_ib_flow_db flow_db; 686 /* protect resources needed as part of reset flow */ 687 spinlock_t reset_flow_resource_lock; 688 struct list_head qp_list; 689 /* Array with num_ports elements */ 690 struct mlx5_ib_port *port; 691 struct mlx5_ib_congestion congestion; 692 }; 693 694 static inline struct mlx5_ib_cq *to_mibcq(struct mlx5_core_cq *mcq) 695 { 696 return container_of(mcq, struct mlx5_ib_cq, mcq); 697 } 698 699 static inline struct mlx5_ib_xrcd *to_mxrcd(struct ib_xrcd *ibxrcd) 700 { 701 return container_of(ibxrcd, struct mlx5_ib_xrcd, ibxrcd); 702 } 703 704 static inline struct mlx5_ib_dev *to_mdev(struct ib_device *ibdev) 705 { 706 return container_of(ibdev, struct mlx5_ib_dev, ib_dev); 707 } 708 709 static inline struct mlx5_ib_cq *to_mcq(struct ib_cq *ibcq) 710 { 711 return container_of(ibcq, struct mlx5_ib_cq, ibcq); 712 } 713 714 static inline struct mlx5_ib_qp *to_mibqp(struct mlx5_core_qp *mqp) 715 { 716 return container_of(mqp, struct mlx5_ib_qp_base, mqp)->container_mibqp; 717 } 718 719 static inline struct mlx5_ib_rwq *to_mibrwq(struct mlx5_core_qp *core_qp) 720 { 721 return container_of(core_qp, struct mlx5_ib_rwq, core_qp); 722 } 723 724 static inline struct mlx5_ib_mr *to_mibmr(struct mlx5_core_mr *mmkey) 725 { 726 return container_of(mmkey, struct mlx5_ib_mr, mmkey); 727 } 728 729 static inline struct mlx5_ib_pd *to_mpd(struct ib_pd *ibpd) 730 { 731 return container_of(ibpd, struct mlx5_ib_pd, ibpd); 732 } 733 734 static inline struct mlx5_ib_srq *to_msrq(struct ib_srq *ibsrq) 735 { 736 return container_of(ibsrq, struct mlx5_ib_srq, ibsrq); 737 } 738 739 static inline struct mlx5_ib_qp *to_mqp(struct ib_qp *ibqp) 740 { 741 return container_of(ibqp, struct mlx5_ib_qp, ibqp); 742 } 743 744 static inline struct mlx5_ib_rwq *to_mrwq(struct ib_wq *ibwq) 745 { 746 return container_of(ibwq, struct mlx5_ib_rwq, ibwq); 747 } 748 749 static inline struct mlx5_ib_rwq_ind_table *to_mrwq_ind_table(struct ib_rwq_ind_table *ib_rwq_ind_tbl) 750 { 751 return container_of(ib_rwq_ind_tbl, struct mlx5_ib_rwq_ind_table, ib_rwq_ind_tbl); 752 } 753 754 static inline struct mlx5_ib_srq *to_mibsrq(struct mlx5_core_srq *msrq) 755 { 756 return container_of(msrq, struct mlx5_ib_srq, msrq); 757 } 758 759 static inline struct mlx5_ib_mr *to_mmr(struct ib_mr *ibmr) 760 { 761 return container_of(ibmr, struct mlx5_ib_mr, ibmr); 762 } 763 764 static inline struct mlx5_ib_mw *to_mmw(struct ib_mw *ibmw) 765 { 766 return container_of(ibmw, struct mlx5_ib_mw, ibmw); 767 } 768 769 struct mlx5_ib_ah { 770 struct ib_ah ibah; 771 struct mlx5_av av; 772 }; 773 774 static inline struct mlx5_ib_ah *to_mah(struct ib_ah *ibah) 775 { 776 return container_of(ibah, struct mlx5_ib_ah, ibah); 777 } 778 779 int mlx5_ib_db_map_user(struct mlx5_ib_ucontext *context, unsigned long virt, 780 struct mlx5_db *db); 781 void mlx5_ib_db_unmap_user(struct mlx5_ib_ucontext *context, struct mlx5_db *db); 782 void __mlx5_ib_cq_clean(struct mlx5_ib_cq *cq, u32 qpn, struct mlx5_ib_srq *srq); 783 void mlx5_ib_cq_clean(struct mlx5_ib_cq *cq, u32 qpn, struct mlx5_ib_srq *srq); 784 void mlx5_ib_free_srq_wqe(struct mlx5_ib_srq *srq, int wqe_index); 785 int mlx5_MAD_IFC(struct mlx5_ib_dev *dev, int ignore_mkey, int ignore_bkey, 786 u8 port, const struct ib_wc *in_wc, const struct ib_grh *in_grh, 787 const void *in_mad, void *response_mad); 788 struct ib_ah *mlx5_ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr, 789 struct ib_udata *udata); 790 int mlx5_ib_query_ah(struct ib_ah *ibah, struct ib_ah_attr *ah_attr); 791 int mlx5_ib_destroy_ah(struct ib_ah *ah); 792 struct ib_srq *mlx5_ib_create_srq(struct ib_pd *pd, 793 struct ib_srq_init_attr *init_attr, 794 struct ib_udata *udata); 795 int mlx5_ib_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr, 796 enum ib_srq_attr_mask attr_mask, struct ib_udata *udata); 797 int mlx5_ib_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *srq_attr); 798 int mlx5_ib_destroy_srq(struct ib_srq *srq); 799 int mlx5_ib_post_srq_recv(struct ib_srq *ibsrq, struct ib_recv_wr *wr, 800 struct ib_recv_wr **bad_wr); 801 struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd, 802 struct ib_qp_init_attr *init_attr, 803 struct ib_udata *udata); 804 int mlx5_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, 805 int attr_mask, struct ib_udata *udata); 806 int mlx5_ib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr, int qp_attr_mask, 807 struct ib_qp_init_attr *qp_init_attr); 808 int mlx5_ib_destroy_qp(struct ib_qp *qp); 809 int mlx5_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr, 810 struct ib_send_wr **bad_wr); 811 int mlx5_ib_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr, 812 struct ib_recv_wr **bad_wr); 813 void *mlx5_get_send_wqe(struct mlx5_ib_qp *qp, int n); 814 int mlx5_ib_read_user_wqe(struct mlx5_ib_qp *qp, int send, int wqe_index, 815 void *buffer, u32 length, 816 struct mlx5_ib_qp_base *base); 817 struct ib_cq *mlx5_ib_create_cq(struct ib_device *ibdev, 818 const struct ib_cq_init_attr *attr, 819 struct ib_ucontext *context, 820 struct ib_udata *udata); 821 int mlx5_ib_destroy_cq(struct ib_cq *cq); 822 int mlx5_ib_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc); 823 int mlx5_ib_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags); 824 int mlx5_ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period); 825 int mlx5_ib_resize_cq(struct ib_cq *ibcq, int entries, struct ib_udata *udata); 826 struct ib_mr *mlx5_ib_get_dma_mr(struct ib_pd *pd, int acc); 827 struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, 828 u64 virt_addr, int access_flags, 829 struct ib_udata *udata); 830 struct ib_mw *mlx5_ib_alloc_mw(struct ib_pd *pd, enum ib_mw_type type, 831 struct ib_udata *udata); 832 int mlx5_ib_dealloc_mw(struct ib_mw *mw); 833 int mlx5_ib_update_mtt(struct mlx5_ib_mr *mr, u64 start_page_index, 834 int npages, int zap); 835 int mlx5_ib_rereg_user_mr(struct ib_mr *ib_mr, int flags, u64 start, 836 u64 length, u64 virt_addr, int access_flags, 837 struct ib_pd *pd, struct ib_udata *udata); 838 int mlx5_ib_dereg_mr(struct ib_mr *ibmr); 839 struct ib_mr *mlx5_ib_alloc_mr(struct ib_pd *pd, 840 enum ib_mr_type mr_type, 841 u32 max_num_sg); 842 int mlx5_ib_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg, int sg_nents, 843 unsigned int *sg_offset); 844 int mlx5_ib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num, 845 const struct ib_wc *in_wc, const struct ib_grh *in_grh, 846 const struct ib_mad_hdr *in, size_t in_mad_size, 847 struct ib_mad_hdr *out, size_t *out_mad_size, 848 u16 *out_mad_pkey_index); 849 struct ib_xrcd *mlx5_ib_alloc_xrcd(struct ib_device *ibdev, 850 struct ib_ucontext *context, 851 struct ib_udata *udata); 852 int mlx5_ib_dealloc_xrcd(struct ib_xrcd *xrcd); 853 int mlx5_ib_get_buf_offset(u64 addr, int page_shift, u32 *offset); 854 int mlx5_query_ext_port_caps(struct mlx5_ib_dev *dev, u8 port); 855 int mlx5_query_mad_ifc_smp_attr_node_info(struct ib_device *ibdev, 856 struct ib_smp *out_mad); 857 int mlx5_query_mad_ifc_system_image_guid(struct ib_device *ibdev, 858 __be64 *sys_image_guid); 859 int mlx5_query_mad_ifc_max_pkeys(struct ib_device *ibdev, 860 u16 *max_pkeys); 861 int mlx5_query_mad_ifc_vendor_id(struct ib_device *ibdev, 862 u32 *vendor_id); 863 int mlx5_query_mad_ifc_node_desc(struct mlx5_ib_dev *dev, char *node_desc); 864 int mlx5_query_mad_ifc_node_guid(struct mlx5_ib_dev *dev, __be64 *node_guid); 865 int mlx5_query_mad_ifc_pkey(struct ib_device *ibdev, u8 port, u16 index, 866 u16 *pkey); 867 int mlx5_query_mad_ifc_gids(struct ib_device *ibdev, u8 port, int index, 868 union ib_gid *gid); 869 int mlx5_query_mad_ifc_port(struct ib_device *ibdev, u8 port, 870 struct ib_port_attr *props); 871 int mlx5_ib_query_port(struct ib_device *ibdev, u8 port, 872 struct ib_port_attr *props); 873 int mlx5_ib_init_fmr(struct mlx5_ib_dev *dev); 874 void mlx5_ib_cleanup_fmr(struct mlx5_ib_dev *dev); 875 void mlx5_ib_cont_pages(struct ib_umem *umem, u64 addr, int *count, int *shift, 876 int *ncont, int *order); 877 void __mlx5_ib_populate_pas(struct mlx5_ib_dev *dev, struct ib_umem *umem, 878 int page_shift, size_t offset, size_t num_pages, 879 __be64 *pas, int access_flags); 880 void mlx5_ib_populate_pas(struct mlx5_ib_dev *dev, struct ib_umem *umem, 881 int page_shift, __be64 *pas, int access_flags); 882 void mlx5_ib_copy_pas(u64 *old, u64 *new, int step, int num); 883 int mlx5_ib_get_cqe_size(struct mlx5_ib_dev *dev, struct ib_cq *ibcq); 884 int mlx5_mr_cache_init(struct mlx5_ib_dev *dev); 885 int mlx5_mr_cache_cleanup(struct mlx5_ib_dev *dev); 886 int mlx5_mr_ib_cont_pages(struct ib_umem *umem, u64 addr, int *count, int *shift); 887 int mlx5_ib_check_mr_status(struct ib_mr *ibmr, u32 check_mask, 888 struct ib_mr_status *mr_status); 889 struct ib_wq *mlx5_ib_create_wq(struct ib_pd *pd, 890 struct ib_wq_init_attr *init_attr, 891 struct ib_udata *udata); 892 int mlx5_ib_destroy_wq(struct ib_wq *wq); 893 int mlx5_ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *wq_attr, 894 u32 wq_attr_mask, struct ib_udata *udata); 895 struct ib_rwq_ind_table *mlx5_ib_create_rwq_ind_table(struct ib_device *device, 896 struct ib_rwq_ind_table_init_attr *init_attr, 897 struct ib_udata *udata); 898 int mlx5_ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table); 899 900 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING 901 extern struct workqueue_struct *mlx5_ib_page_fault_wq; 902 903 void mlx5_ib_internal_fill_odp_caps(struct mlx5_ib_dev *dev); 904 void mlx5_ib_mr_pfault_handler(struct mlx5_ib_qp *qp, 905 struct mlx5_ib_pfault *pfault); 906 void mlx5_ib_odp_create_qp(struct mlx5_ib_qp *qp); 907 int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev); 908 void mlx5_ib_odp_remove_one(struct mlx5_ib_dev *ibdev); 909 int __init mlx5_ib_odp_init(void); 910 void mlx5_ib_odp_cleanup(void); 911 void mlx5_ib_qp_disable_pagefaults(struct mlx5_ib_qp *qp); 912 void mlx5_ib_qp_enable_pagefaults(struct mlx5_ib_qp *qp); 913 void mlx5_ib_invalidate_range(struct ib_umem *umem, unsigned long start, 914 unsigned long end); 915 #else /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */ 916 static inline void mlx5_ib_internal_fill_odp_caps(struct mlx5_ib_dev *dev) 917 { 918 return; 919 } 920 921 static inline void mlx5_ib_odp_create_qp(struct mlx5_ib_qp *qp) {} 922 static inline int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev) { return 0; } 923 static inline void mlx5_ib_odp_remove_one(struct mlx5_ib_dev *ibdev) {} 924 static inline int mlx5_ib_odp_init(void) { return 0; } 925 static inline void mlx5_ib_odp_cleanup(void) {} 926 static inline void mlx5_ib_qp_disable_pagefaults(struct mlx5_ib_qp *qp) {} 927 static inline void mlx5_ib_qp_enable_pagefaults(struct mlx5_ib_qp *qp) {} 928 929 #endif /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */ 930 931 int mlx5_ib_get_vf_config(struct ib_device *device, int vf, 932 u8 port, struct ifla_vf_info *info); 933 int mlx5_ib_set_vf_link_state(struct ib_device *device, int vf, 934 u8 port, int state); 935 int mlx5_ib_get_vf_stats(struct ib_device *device, int vf, 936 u8 port, struct ifla_vf_stats *stats); 937 int mlx5_ib_set_vf_guid(struct ib_device *device, int vf, u8 port, 938 u64 guid, int type); 939 940 __be16 mlx5_get_roce_udp_sport(struct mlx5_ib_dev *dev, u8 port_num, 941 int index); 942 int mlx5_get_roce_gid_type(struct mlx5_ib_dev *dev, u8 port_num, 943 int index, enum ib_gid_type *gid_type); 944 945 /* GSI QP helper functions */ 946 struct ib_qp *mlx5_ib_gsi_create_qp(struct ib_pd *pd, 947 struct ib_qp_init_attr *init_attr); 948 int mlx5_ib_gsi_destroy_qp(struct ib_qp *qp); 949 int mlx5_ib_gsi_modify_qp(struct ib_qp *qp, struct ib_qp_attr *attr, 950 int attr_mask); 951 int mlx5_ib_gsi_query_qp(struct ib_qp *qp, struct ib_qp_attr *qp_attr, 952 int qp_attr_mask, 953 struct ib_qp_init_attr *qp_init_attr); 954 int mlx5_ib_gsi_post_send(struct ib_qp *qp, struct ib_send_wr *wr, 955 struct ib_send_wr **bad_wr); 956 int mlx5_ib_gsi_post_recv(struct ib_qp *qp, struct ib_recv_wr *wr, 957 struct ib_recv_wr **bad_wr); 958 void mlx5_ib_gsi_pkey_change(struct mlx5_ib_gsi_qp *gsi); 959 960 int mlx5_ib_generate_wc(struct ib_cq *ibcq, struct ib_wc *wc); 961 962 static inline void init_query_mad(struct ib_smp *mad) 963 { 964 mad->base_version = 1; 965 mad->mgmt_class = IB_MGMT_CLASS_SUBN_LID_ROUTED; 966 mad->class_version = 1; 967 mad->method = IB_MGMT_METHOD_GET; 968 } 969 970 static inline u8 convert_access(int acc) 971 { 972 return (acc & IB_ACCESS_REMOTE_ATOMIC ? MLX5_PERM_ATOMIC : 0) | 973 (acc & IB_ACCESS_REMOTE_WRITE ? MLX5_PERM_REMOTE_WRITE : 0) | 974 (acc & IB_ACCESS_REMOTE_READ ? MLX5_PERM_REMOTE_READ : 0) | 975 (acc & IB_ACCESS_LOCAL_WRITE ? MLX5_PERM_LOCAL_WRITE : 0) | 976 MLX5_PERM_LOCAL_READ; 977 } 978 979 static inline int is_qp1(enum ib_qp_type qp_type) 980 { 981 return qp_type == MLX5_IB_QPT_HW_GSI; 982 } 983 984 #define MLX5_MAX_UMR_SHIFT 16 985 #define MLX5_MAX_UMR_PAGES (1 << MLX5_MAX_UMR_SHIFT) 986 987 static inline u32 check_cq_create_flags(u32 flags) 988 { 989 /* 990 * It returns non-zero value for unsupported CQ 991 * create flags, otherwise it returns zero. 992 */ 993 return (flags & ~(IB_CQ_FLAGS_IGNORE_OVERRUN | 994 IB_CQ_FLAGS_TIMESTAMP_COMPLETION)); 995 } 996 997 static inline int verify_assign_uidx(u8 cqe_version, u32 cmd_uidx, 998 u32 *user_index) 999 { 1000 if (cqe_version) { 1001 if ((cmd_uidx == MLX5_IB_DEFAULT_UIDX) || 1002 (cmd_uidx & ~MLX5_USER_ASSIGNED_UIDX_MASK)) 1003 return -EINVAL; 1004 *user_index = cmd_uidx; 1005 } else { 1006 *user_index = MLX5_IB_DEFAULT_UIDX; 1007 } 1008 1009 return 0; 1010 } 1011 1012 static inline int get_qp_user_index(struct mlx5_ib_ucontext *ucontext, 1013 struct mlx5_ib_create_qp *ucmd, 1014 int inlen, 1015 u32 *user_index) 1016 { 1017 u8 cqe_version = ucontext->cqe_version; 1018 1019 if (field_avail(struct mlx5_ib_create_qp, uidx, inlen) && 1020 !cqe_version && (ucmd->uidx == MLX5_IB_DEFAULT_UIDX)) 1021 return 0; 1022 1023 if (!!(field_avail(struct mlx5_ib_create_qp, uidx, inlen) != 1024 !!cqe_version)) 1025 return -EINVAL; 1026 1027 return verify_assign_uidx(cqe_version, ucmd->uidx, user_index); 1028 } 1029 1030 static inline int get_srq_user_index(struct mlx5_ib_ucontext *ucontext, 1031 struct mlx5_ib_create_srq *ucmd, 1032 int inlen, 1033 u32 *user_index) 1034 { 1035 u8 cqe_version = ucontext->cqe_version; 1036 1037 if (field_avail(struct mlx5_ib_create_srq, uidx, inlen) && 1038 !cqe_version && (ucmd->uidx == MLX5_IB_DEFAULT_UIDX)) 1039 return 0; 1040 1041 if (!!(field_avail(struct mlx5_ib_create_srq, uidx, inlen) != 1042 !!cqe_version)) 1043 return -EINVAL; 1044 1045 return verify_assign_uidx(cqe_version, ucmd->uidx, user_index); 1046 } 1047 1048 void mlx5_ib_cleanup_congestion(struct mlx5_ib_dev *); 1049 int mlx5_ib_init_congestion(struct mlx5_ib_dev *); 1050 1051 #endif /* MLX5_IB_H */ 1052