1#ifndef DEF_RDMA_VT_H 2#define DEF_RDMA_VT_H 3 4/*- 5 * SPDX-License-Identifier: BSD-2-Clause OR GPL-2.0 6 * 7 * Copyright(c) 2016 Intel Corporation. 8 * 9 * This file is provided under a dual BSD/GPLv2 license. When using or 10 * redistributing this file, you may do so under either license. 11 * 12 * GPL LICENSE SUMMARY 13 * 14 * This program is free software; you can redistribute it and/or modify 15 * it under the terms of version 2 of the GNU General Public License as 16 * published by the Free Software Foundation. 17 * 18 * This program is distributed in the hope that it will be useful, but 19 * WITHOUT ANY WARRANTY; without even the implied warranty of 20 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 21 * General Public License for more details. 22 * 23 * BSD LICENSE 24 * 25 * Redistribution and use in source and binary forms, with or without 26 * modification, are permitted provided that the following conditions 27 * are met: 28 * 29 * - Redistributions of source code must retain the above copyright 30 * notice, this list of conditions and the following disclaimer. 31 * - Redistributions in binary form must reproduce the above copyright 32 * notice, this list of conditions and the following disclaimer in 33 * the documentation and/or other materials provided with the 34 * distribution. 35 * - Neither the name of Intel Corporation nor the names of its 36 * contributors may be used to endorse or promote products derived 37 * from this software without specific prior written permission. 38 * 39 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 40 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 41 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 42 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 43 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 44 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 45 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 46 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 47 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 48 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 49 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 50 * 51 * $FreeBSD$ 52 */ 53 54/* 55 * Structure that low level drivers will populate in order to register with the 56 * rdmavt layer. 57 */ 58 59#include <linux/spinlock.h> 60#include <linux/list.h> 61#include <linux/hash.h> 62#include <rdma/ib_verbs.h> 63#include <rdma/rdmavt_mr.h> 64#include <rdma/rdmavt_qp.h> 65 66#define RVT_MAX_PKEY_VALUES 16 67 68struct rvt_ibport { 69 struct rvt_qp __rcu *qp[2]; 70 struct ib_mad_agent *send_agent; /* agent for SMI (traps) */ 71 struct rb_root mcast_tree; 72 spinlock_t lock; /* protect changes in this struct */ 73 74 /* non-zero when timer is set */ 75 unsigned long mkey_lease_timeout; 76 unsigned long trap_timeout; 77 __be64 gid_prefix; /* in network order */ 78 __be64 mkey; 79 u64 tid; 80 u32 port_cap_flags; 81 u32 pma_sample_start; 82 u32 pma_sample_interval; 83 __be16 pma_counter_select[5]; 84 u16 pma_tag; 85 u16 mkey_lease_period; 86 u16 sm_lid; 87 u8 sm_sl; 88 u8 mkeyprot; 89 u8 subnet_timeout; 90 u8 vl_high_limit; 91 92 /* 93 * Driver is expected to keep these up to date. These 94 * counters are informational only and not required to be 95 * completely accurate. 96 */ 97 u64 n_rc_resends; 98 u64 n_seq_naks; 99 u64 n_rdma_seq; 100 u64 n_rnr_naks; 101 u64 n_other_naks; 102 u64 n_loop_pkts; 103 u64 n_pkt_drops; 104 u64 n_vl15_dropped; 105 u64 n_rc_timeouts; 106 u64 n_dmawait; 107 u64 n_unaligned; 108 u64 n_rc_dupreq; 109 u64 n_rc_seqnak; 110 u16 pkey_violations; 111 u16 qkey_violations; 112 u16 mkey_violations; 113 114 /* Hot-path per CPU counters to avoid cacheline trading to update */ 115 u64 z_rc_acks; 116 u64 z_rc_qacks; 117 u64 z_rc_delayed_comp; 118 u64 __percpu *rc_acks; 119 u64 __percpu *rc_qacks; 120 u64 __percpu *rc_delayed_comp; 121 122 void *priv; /* driver private data */ 123 124 /* 125 * The pkey table is allocated and maintained by the driver. Drivers 126 * need to have access to this before registering with rdmav. However 127 * rdmavt will need access to it so drivers need to proviee this during 128 * the attach port API call. 129 */ 130 u16 *pkey_table; 131 132 struct rvt_ah *sm_ah; 133}; 134 135#define RVT_CQN_MAX 16 /* maximum length of cq name */ 136 137/* 138 * Things that are driver specific, module parameters in hfi1 and qib 139 */ 140struct rvt_driver_params { 141 struct ib_device_attr props; 142 143 /* 144 * Anything driver specific that is not covered by props 145 * For instance special module parameters. Goes here. 146 */ 147 unsigned int lkey_table_size; 148 unsigned int qp_table_size; 149 int qpn_start; 150 int qpn_inc; 151 int qpn_res_start; 152 int qpn_res_end; 153 int nports; 154 int npkeys; 155 char cq_name[RVT_CQN_MAX]; 156 int node; 157 int psn_mask; 158 int psn_shift; 159 int psn_modify_mask; 160 u32 core_cap_flags; 161 u32 max_mad_size; 162 u8 qos_shift; 163 u8 max_rdma_atomic; 164 u8 reserved_operations; 165}; 166 167/* Protection domain */ 168struct rvt_pd { 169 struct ib_pd ibpd; 170 int user; /* non-zero if created from user space */ 171}; 172 173/* Address handle */ 174struct rvt_ah { 175 struct ib_ah ibah; 176 struct ib_ah_attr attr; 177 atomic_t refcount; 178 u8 vl; 179 u8 log_pmtu; 180}; 181 182struct rvt_dev_info; 183struct rvt_swqe; 184struct rvt_driver_provided { 185 /* 186 * Which functions are required depends on which verbs rdmavt is 187 * providing and which verbs the driver is overriding. See 188 * check_support() for details. 189 */ 190 191 /* Passed to ib core registration. Callback to create syfs files */ 192 int (*port_callback)(struct ib_device *, u8, struct kobject *); 193 194 /* 195 * Returns a string to represent the device for which is being 196 * registered. This is primarily used for error and debug messages on 197 * the console. 198 */ 199 const char * (*get_card_name)(struct rvt_dev_info *rdi); 200 201 /* 202 * Returns a pointer to the undelying hardware's PCI device. This is 203 * used to display information as to what hardware is being referenced 204 * in an output message 205 */ 206 struct pci_dev * (*get_pci_dev)(struct rvt_dev_info *rdi); 207 208 /* 209 * Allocate a private queue pair data structure for driver specific 210 * information which is opaque to rdmavt. Errors are returned via 211 * ERR_PTR(err). The driver is free to return NULL or a valid 212 * pointer. 213 */ 214 void * (*qp_priv_alloc)(struct rvt_dev_info *rdi, struct rvt_qp *qp, 215 gfp_t gfp); 216 217 /* 218 * Free the driver's private qp structure. 219 */ 220 void (*qp_priv_free)(struct rvt_dev_info *rdi, struct rvt_qp *qp); 221 222 /* 223 * Inform the driver the particular qp in quesiton has been reset so 224 * that it can clean up anything it needs to. 225 */ 226 void (*notify_qp_reset)(struct rvt_qp *qp); 227 228 /* 229 * Give the driver a notice that there is send work to do. It is up to 230 * the driver to generally push the packets out, this just queues the 231 * work with the driver. There are two variants here. The no_lock 232 * version requires the s_lock not to be held. The other assumes the 233 * s_lock is held. 234 */ 235 void (*schedule_send)(struct rvt_qp *qp); 236 void (*schedule_send_no_lock)(struct rvt_qp *qp); 237 238 /* 239 * Sometimes rdmavt needs to kick the driver's send progress. That is 240 * done by this call back. 241 */ 242 void (*do_send)(struct rvt_qp *qp); 243 244 /* 245 * Get a path mtu from the driver based on qp attributes. 246 */ 247 int (*get_pmtu_from_attr)(struct rvt_dev_info *rdi, struct rvt_qp *qp, 248 struct ib_qp_attr *attr); 249 250 /* 251 * Notify driver that it needs to flush any outstanding IO requests that 252 * are waiting on a qp. 253 */ 254 void (*flush_qp_waiters)(struct rvt_qp *qp); 255 256 /* 257 * Notify driver to stop its queue of sending packets. Nothing else 258 * should be posted to the queue pair after this has been called. 259 */ 260 void (*stop_send_queue)(struct rvt_qp *qp); 261 262 /* 263 * Have the drivr drain any in progress operations 264 */ 265 void (*quiesce_qp)(struct rvt_qp *qp); 266 267 /* 268 * Inform the driver a qp has went to error state. 269 */ 270 void (*notify_error_qp)(struct rvt_qp *qp); 271 272 /* 273 * Get an MTU for a qp. 274 */ 275 u32 (*mtu_from_qp)(struct rvt_dev_info *rdi, struct rvt_qp *qp, 276 u32 pmtu); 277 /* 278 * Convert an mtu to a path mtu 279 */ 280 int (*mtu_to_path_mtu)(u32 mtu); 281 282 /* 283 * Get the guid of a port in big endian byte order 284 */ 285 int (*get_guid_be)(struct rvt_dev_info *rdi, struct rvt_ibport *rvp, 286 int guid_index, __be64 *guid); 287 288 /* 289 * Query driver for the state of the port. 290 */ 291 int (*query_port_state)(struct rvt_dev_info *rdi, u8 port_num, 292 struct ib_port_attr *props); 293 294 /* 295 * Tell driver to shutdown a port 296 */ 297 int (*shut_down_port)(struct rvt_dev_info *rdi, u8 port_num); 298 299 /* Tell driver to send a trap for changed port capabilities */ 300 void (*cap_mask_chg)(struct rvt_dev_info *rdi, u8 port_num); 301 302 /* 303 * The following functions can be safely ignored completely. Any use of 304 * these is checked for NULL before blindly calling. Rdmavt should also 305 * be functional if drivers omit these. 306 */ 307 308 /* Called to inform the driver that all qps should now be freed. */ 309 unsigned (*free_all_qps)(struct rvt_dev_info *rdi); 310 311 /* Driver specific AH validation */ 312 int (*check_ah)(struct ib_device *, struct ib_ah_attr *); 313 314 /* Inform the driver a new AH has been created */ 315 void (*notify_new_ah)(struct ib_device *, struct ib_ah_attr *, 316 struct rvt_ah *); 317 318 /* Let the driver pick the next queue pair number*/ 319 int (*alloc_qpn)(struct rvt_dev_info *rdi, struct rvt_qpn_table *qpt, 320 enum ib_qp_type type, u8 port_num, gfp_t gfp); 321 322 /* Determine if its safe or allowed to modify the qp */ 323 int (*check_modify_qp)(struct rvt_qp *qp, struct ib_qp_attr *attr, 324 int attr_mask, struct ib_udata *udata); 325 326 /* Driver specific QP modification/notification-of */ 327 void (*modify_qp)(struct rvt_qp *qp, struct ib_qp_attr *attr, 328 int attr_mask, struct ib_udata *udata); 329 330 /* Driver specific work request checking */ 331 int (*check_send_wqe)(struct rvt_qp *qp, struct rvt_swqe *wqe); 332 333 /* Notify driver a mad agent has been created */ 334 void (*notify_create_mad_agent)(struct rvt_dev_info *rdi, int port_idx); 335 336 /* Notify driver a mad agent has been removed */ 337 void (*notify_free_mad_agent)(struct rvt_dev_info *rdi, int port_idx); 338 339}; 340 341struct rvt_dev_info { 342 struct ib_device ibdev; /* Keep this first. Nothing above here */ 343 344 /* 345 * Prior to calling for registration the driver will be responsible for 346 * allocating space for this structure. 347 * 348 * The driver will also be responsible for filling in certain members of 349 * dparms.props. The driver needs to fill in dparms exactly as it would 350 * want values reported to a ULP. This will be returned to the caller 351 * in rdmavt's device. The driver should also therefore refrain from 352 * modifying this directly after registration with rdmavt. 353 */ 354 355 /* Driver specific properties */ 356 struct rvt_driver_params dparms; 357 358 /* post send table */ 359 const struct rvt_operation_params *post_parms; 360 361 struct rvt_mregion __rcu *dma_mr; 362 struct rvt_lkey_table lkey_table; 363 364 /* Driver specific helper functions */ 365 struct rvt_driver_provided driver_f; 366 367 /* Internal use */ 368 int n_pds_allocated; 369 spinlock_t n_pds_lock; /* Protect pd allocated count */ 370 371 int n_ahs_allocated; 372 spinlock_t n_ahs_lock; /* Protect ah allocated count */ 373 374 u32 n_srqs_allocated; 375 spinlock_t n_srqs_lock; /* Protect srqs allocated count */ 376 377 int flags; 378 struct rvt_ibport **ports; 379 380 /* QP */ 381 struct rvt_qp_ibdev *qp_dev; 382 u32 n_qps_allocated; /* number of QPs allocated for device */ 383 u32 n_rc_qps; /* number of RC QPs allocated for device */ 384 u32 busy_jiffies; /* timeout scaling based on RC QP count */ 385 spinlock_t n_qps_lock; /* protect qps, rc qps and busy jiffy counts */ 386 387 /* memory maps */ 388 struct list_head pending_mmaps; 389 spinlock_t mmap_offset_lock; /* protect mmap_offset */ 390 u32 mmap_offset; 391 spinlock_t pending_lock; /* protect pending mmap list */ 392 393 /* CQ */ 394 struct kthread_worker *worker; /* per device cq worker */ 395 u32 n_cqs_allocated; /* number of CQs allocated for device */ 396 spinlock_t n_cqs_lock; /* protect count of in use cqs */ 397 398 /* Multicast */ 399 u32 n_mcast_grps_allocated; /* number of mcast groups allocated */ 400 spinlock_t n_mcast_grps_lock; 401 402}; 403 404static inline struct rvt_pd *ibpd_to_rvtpd(struct ib_pd *ibpd) 405{ 406 return container_of(ibpd, struct rvt_pd, ibpd); 407} 408 409static inline struct rvt_ah *ibah_to_rvtah(struct ib_ah *ibah) 410{ 411 return container_of(ibah, struct rvt_ah, ibah); 412} 413 414static inline struct rvt_dev_info *ib_to_rvt(struct ib_device *ibdev) 415{ 416 return container_of(ibdev, struct rvt_dev_info, ibdev); 417} 418 419static inline struct rvt_srq *ibsrq_to_rvtsrq(struct ib_srq *ibsrq) 420{ 421 return container_of(ibsrq, struct rvt_srq, ibsrq); 422} 423 424static inline struct rvt_qp *ibqp_to_rvtqp(struct ib_qp *ibqp) 425{ 426 return container_of(ibqp, struct rvt_qp, ibqp); 427} 428 429static inline unsigned rvt_get_npkeys(struct rvt_dev_info *rdi) 430{ 431 /* 432 * All ports have same number of pkeys. 433 */ 434 return rdi->dparms.npkeys; 435} 436 437/* 438 * Return the max atomic suitable for determining 439 * the size of the ack ring buffer in a QP. 440 */ 441static inline unsigned int rvt_max_atomic(struct rvt_dev_info *rdi) 442{ 443 return rdi->dparms.max_rdma_atomic + 1; 444} 445 446/* 447 * Return the indexed PKEY from the port PKEY table. 448 */ 449static inline u16 rvt_get_pkey(struct rvt_dev_info *rdi, 450 int port_index, 451 unsigned index) 452{ 453 if (index >= rvt_get_npkeys(rdi)) 454 return 0; 455 else 456 return rdi->ports[port_index]->pkey_table[index]; 457} 458 459/** 460 * rvt_lookup_qpn - return the QP with the given QPN 461 * @ibp: the ibport 462 * @qpn: the QP number to look up 463 * 464 * The caller must hold the rcu_read_lock(), and keep the lock until 465 * the returned qp is no longer in use. 466 */ 467/* TODO: Remove this and put in rdmavt/qp.h when no longer needed by drivers */ 468static inline struct rvt_qp *rvt_lookup_qpn(struct rvt_dev_info *rdi, 469 struct rvt_ibport *rvp, 470 u32 qpn) __must_hold(RCU) 471{ 472 struct rvt_qp *qp = NULL; 473 474 if (unlikely(qpn <= 1)) { 475 qp = rcu_dereference(rvp->qp[qpn]); 476 } else { 477 u32 n = hash_32(qpn, rdi->qp_dev->qp_table_bits); 478 479 for (qp = rcu_dereference(rdi->qp_dev->qp_table[n]); qp; 480 qp = rcu_dereference(qp->next)) 481 if (qp->ibqp.qp_num == qpn) 482 break; 483 } 484 return qp; 485} 486 487struct rvt_dev_info *rvt_alloc_device(size_t size, int nports); 488void rvt_dealloc_device(struct rvt_dev_info *rdi); 489int rvt_register_device(struct rvt_dev_info *rvd); 490void rvt_unregister_device(struct rvt_dev_info *rvd); 491int rvt_check_ah(struct ib_device *ibdev, struct ib_ah_attr *ah_attr); 492int rvt_init_port(struct rvt_dev_info *rdi, struct rvt_ibport *port, 493 int port_index, u16 *pkey_table); 494int rvt_fast_reg_mr(struct rvt_qp *qp, struct ib_mr *ibmr, u32 key, 495 int access); 496int rvt_invalidate_rkey(struct rvt_qp *qp, u32 rkey); 497int rvt_rkey_ok(struct rvt_qp *qp, struct rvt_sge *sge, 498 u32 len, u64 vaddr, u32 rkey, int acc); 499int rvt_lkey_ok(struct rvt_lkey_table *rkt, struct rvt_pd *pd, 500 struct rvt_sge *isge, struct ib_sge *sge, int acc); 501struct rvt_mcast *rvt_mcast_find(struct rvt_ibport *ibp, union ib_gid *mgid); 502 503#endif /* DEF_RDMA_VT_H */ 504