include/rdma/rdma_vt.h

   1 #ifndef DEF_RDMA_VT_H
   2 #define DEF_RDMA_VT_H
   3
   4 /*
   5  * Copyright(c) 2016 Intel Corporation.
   6  *
   7  * This file is provided under a dual BSD/GPLv2 license.  When using or
   8  * redistributing this file, you may do so under either license.
   9  *
  10  * GPL LICENSE SUMMARY
  11  *
  12  * This program is free software; you can redistribute it and/or modify
  13  * it under the terms of version 2 of the GNU General Public License as
  14  * published by the Free Software Foundation.
  15  *
  16  * This program is distributed in the hope that it will be useful, but
  17  * WITHOUT ANY WARRANTY; without even the implied warranty of
  18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  19  * General Public License for more details.
  20  *
  21  * BSD LICENSE
  22  *
  23  * Redistribution and use in source and binary forms, with or without
  24  * modification, are permitted provided that the following conditions
  25  * are met:
  26  *
  27  *  - Redistributions of source code must retain the above copyright
  28  *    notice, this list of conditions and the following disclaimer.
  29  *  - Redistributions in binary form must reproduce the above copyright
  30  *    notice, this list of conditions and the following disclaimer in
  31  *    the documentation and/or other materials provided with the
  32  *    distribution.
  33  *  - Neither the name of Intel Corporation nor the names of its
  34  *    contributors may be used to endorse or promote products derived
  35  *    from this software without specific prior written permission.
  36  *
  37  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  38  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  39  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  40  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  41  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  42  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  43  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  44  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  45  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  46  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  47  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  48  *
  49  */
  50
  51 /*
  52  * Structure that low level drivers will populate in order to register with the
  53  * rdmavt layer.
  54  */
  55
  56 #include <linux/spinlock.h>
  57 #include <linux/list.h>
  58 #include <linux/hash.h>
  59 #include <rdma/ib_verbs.h>
  60 #include <rdma/ib_mad.h>
  61 #include <rdma/rdmavt_mr.h>
  62 #include <rdma/rdmavt_qp.h>
  63
  64 #define RVT_MAX_PKEY_VALUES 16
  65
  66 #define RVT_MAX_TRAP_LEN 100 /* Limit pending trap list */
  67 #define RVT_MAX_TRAP_LISTS 5 /*((IB_NOTICE_TYPE_INFO & 0x0F) + 1)*/
  68 #define RVT_TRAP_TIMEOUT 4096 /* 4.096 usec */
  69
  70 struct trap_list {
  71         u32 list_len;
  72         struct list_head list;
  73 };
  74
  75 struct rvt_ibport {
  76         struct rvt_qp __rcu *qp[2];
  77         struct ib_mad_agent *send_agent;        /* agent for SMI (traps) */
  78         struct rb_root mcast_tree;
  79         spinlock_t lock;                /* protect changes in this struct */
  80
  81         /* non-zero when timer is set */
  82         unsigned long mkey_lease_timeout;
  83         unsigned long trap_timeout;
  84         __be64 gid_prefix;      /* in network order */
  85         __be64 mkey;
  86         u64 tid;
  87         u32 port_cap_flags;
  88         u16 port_cap3_flags;
  89         u32 pma_sample_start;
  90         u32 pma_sample_interval;
  91         __be16 pma_counter_select[5];
  92         u16 pma_tag;
  93         u16 mkey_lease_period;
  94         u32 sm_lid;
  95         u8 sm_sl;
  96         u8 mkeyprot;
  97         u8 subnet_timeout;
  98         u8 vl_high_limit;
  99
 100         /*
 101          * Driver is expected to keep these up to date. These
 102          * counters are informational only and not required to be
 103          * completely accurate.
 104          */
 105         u64 n_rc_resends;
 106         u64 n_seq_naks;
 107         u64 n_rdma_seq;
 108         u64 n_rnr_naks;
 109         u64 n_other_naks;
 110         u64 n_loop_pkts;
 111         u64 n_pkt_drops;
 112         u64 n_vl15_dropped;
 113         u64 n_rc_timeouts;
 114         u64 n_dmawait;
 115         u64 n_unaligned;
 116         u64 n_rc_dupreq;
 117         u64 n_rc_seqnak;
 118         u16 pkey_violations;
 119         u16 qkey_violations;
 120         u16 mkey_violations;
 121
 122         /* Hot-path per CPU counters to avoid cacheline trading to update */
 123         u64 z_rc_acks;
 124         u64 z_rc_qacks;
 125         u64 z_rc_delayed_comp;
 126         u64 __percpu *rc_acks;
 127         u64 __percpu *rc_qacks;
 128         u64 __percpu *rc_delayed_comp;
 129
 130         void *priv; /* driver private data */
 131
 132         /*
 133          * The pkey table is allocated and maintained by the driver. Drivers
 134          * need to have access to this before registering with rdmav. However
 135          * rdmavt will need access to it so drivers need to proviee this during
 136          * the attach port API call.
 137          */
 138         u16 *pkey_table;
 139
 140         struct rvt_ah *sm_ah;
 141
 142         /*
 143          * Keep a list of traps that have not been repressed.  They will be
 144          * resent based on trap_timer.
 145          */
 146         struct trap_list trap_lists[RVT_MAX_TRAP_LISTS];
 147         struct timer_list trap_timer;
 148 };
 149
 150 #define RVT_CQN_MAX 16 /* maximum length of cq name */
 151
 152 /*
 153  * Things that are driver specific, module parameters in hfi1 and qib
 154  */
 155 struct rvt_driver_params {
 156         struct ib_device_attr props;
 157
 158         /*
 159          * Anything driver specific that is not covered by props
 160          * For instance special module parameters. Goes here.
 161          */
 162         unsigned int lkey_table_size;
 163         unsigned int qp_table_size;
 164         int qpn_start;
 165         int qpn_inc;
 166         int qpn_res_start;
 167         int qpn_res_end;
 168         int nports;
 169         int npkeys;
 170         char cq_name[RVT_CQN_MAX];
 171         int node;
 172         int psn_mask;
 173         int psn_shift;
 174         int psn_modify_mask;
 175         u32 core_cap_flags;
 176         u32 max_mad_size;
 177         u8 qos_shift;
 178         u8 max_rdma_atomic;
 179         u8 reserved_operations;
 180 };
 181
 182 /* Protection domain */
 183 struct rvt_pd {
 184         struct ib_pd ibpd;
 185         bool user;
 186 };
 187
 188 /* Address handle */
 189 struct rvt_ah {
 190         struct ib_ah ibah;
 191         struct rdma_ah_attr attr;
 192         atomic_t refcount;
 193         u8 vl;
 194         u8 log_pmtu;
 195 };
 196
 197 struct rvt_dev_info;
 198 struct rvt_swqe;
 199 struct rvt_driver_provided {
 200         /*
 201          * Which functions are required depends on which verbs rdmavt is
 202          * providing and which verbs the driver is overriding. See
 203          * check_support() for details.
 204          */
 205
 206         /* hot path calldowns in a single cacheline */
 207
 208         /*
 209          * Give the driver a notice that there is send work to do. It is up to
 210          * the driver to generally push the packets out, this just queues the
 211          * work with the driver. There are two variants here. The no_lock
 212          * version requires the s_lock not to be held. The other assumes the
 213          * s_lock is held.
 214          */
 215         void (*schedule_send)(struct rvt_qp *qp);
 216         void (*schedule_send_no_lock)(struct rvt_qp *qp);
 217
 218         /* Driver specific work request checking */
 219         int (*check_send_wqe)(struct rvt_qp *qp, struct rvt_swqe *wqe);
 220
 221         /*
 222          * Sometimes rdmavt needs to kick the driver's send progress. That is
 223          * done by this call back.
 224          */
 225         void (*do_send)(struct rvt_qp *qp);
 226
 227         /* Passed to ib core registration. Callback to create syfs files */
 228         int (*port_callback)(struct ib_device *, u8, struct kobject *);
 229
 230         /*
 231          * Returns a string to represent the device for which is being
 232          * registered. This is primarily used for error and debug messages on
 233          * the console.
 234          */
 235         const char * (*get_card_name)(struct rvt_dev_info *rdi);
 236
 237         /*
 238          * Returns a pointer to the undelying hardware's PCI device. This is
 239          * used to display information as to what hardware is being referenced
 240          * in an output message
 241          */
 242         struct pci_dev * (*get_pci_dev)(struct rvt_dev_info *rdi);
 243
 244         /*
 245          * Allocate a private queue pair data structure for driver specific
 246          * information which is opaque to rdmavt.  Errors are returned via
 247          * ERR_PTR(err).  The driver is free to return NULL or a valid
 248          * pointer.
 249          */
 250         void * (*qp_priv_alloc)(struct rvt_dev_info *rdi, struct rvt_qp *qp);
 251
 252         /*
 253          * Free the driver's private qp structure.
 254          */
 255         void (*qp_priv_free)(struct rvt_dev_info *rdi, struct rvt_qp *qp);
 256
 257         /*
 258          * Inform the driver the particular qp in quesiton has been reset so
 259          * that it can clean up anything it needs to.
 260          */
 261         void (*notify_qp_reset)(struct rvt_qp *qp);
 262
 263         /*
 264          * Get a path mtu from the driver based on qp attributes.
 265          */
 266         int (*get_pmtu_from_attr)(struct rvt_dev_info *rdi, struct rvt_qp *qp,
 267                                   struct ib_qp_attr *attr);
 268
 269         /*
 270          * Notify driver that it needs to flush any outstanding IO requests that
 271          * are waiting on a qp.
 272          */
 273         void (*flush_qp_waiters)(struct rvt_qp *qp);
 274
 275         /*
 276          * Notify driver to stop its queue of sending packets. Nothing else
 277          * should be posted to the queue pair after this has been called.
 278          */
 279         void (*stop_send_queue)(struct rvt_qp *qp);
 280
 281         /*
 282          * Have the drivr drain any in progress operations
 283          */
 284         void (*quiesce_qp)(struct rvt_qp *qp);
 285
 286         /*
 287          * Inform the driver a qp has went to error state.
 288          */
 289         void (*notify_error_qp)(struct rvt_qp *qp);
 290
 291         /*
 292          * Get an MTU for a qp.
 293          */
 294         u32 (*mtu_from_qp)(struct rvt_dev_info *rdi, struct rvt_qp *qp,
 295                            u32 pmtu);
 296         /*
 297          * Convert an mtu to a path mtu
 298          */
 299         int (*mtu_to_path_mtu)(u32 mtu);
 300
 301         /*
 302          * Get the guid of a port in big endian byte order
 303          */
 304         int (*get_guid_be)(struct rvt_dev_info *rdi, struct rvt_ibport *rvp,
 305                            int guid_index, __be64 *guid);
 306
 307         /*
 308          * Query driver for the state of the port.
 309          */
 310         int (*query_port_state)(struct rvt_dev_info *rdi, u8 port_num,
 311                                 struct ib_port_attr *props);
 312
 313         /*
 314          * Tell driver to shutdown a port
 315          */
 316         int (*shut_down_port)(struct rvt_dev_info *rdi, u8 port_num);
 317
 318         /* Tell driver to send a trap for changed  port capabilities */
 319         void (*cap_mask_chg)(struct rvt_dev_info *rdi, u8 port_num);
 320
 321         /*
 322          * The following functions can be safely ignored completely. Any use of
 323          * these is checked for NULL before blindly calling. Rdmavt should also
 324          * be functional if drivers omit these.
 325          */
 326
 327         /* Called to inform the driver that all qps should now be freed. */
 328         unsigned (*free_all_qps)(struct rvt_dev_info *rdi);
 329
 330         /* Driver specific AH validation */
 331         int (*check_ah)(struct ib_device *, struct rdma_ah_attr *);
 332
 333         /* Inform the driver a new AH has been created */
 334         void (*notify_new_ah)(struct ib_device *, struct rdma_ah_attr *,
 335                               struct rvt_ah *);
 336
 337         /* Let the driver pick the next queue pair number*/
 338         int (*alloc_qpn)(struct rvt_dev_info *rdi, struct rvt_qpn_table *qpt,
 339                          enum ib_qp_type type, u8 port_num);
 340
 341         /* Determine if its safe or allowed to modify the qp */
 342         int (*check_modify_qp)(struct rvt_qp *qp, struct ib_qp_attr *attr,
 343                                int attr_mask, struct ib_udata *udata);
 344
 345         /* Driver specific QP modification/notification-of */
 346         void (*modify_qp)(struct rvt_qp *qp, struct ib_qp_attr *attr,
 347                           int attr_mask, struct ib_udata *udata);
 348
 349         /* Notify driver a mad agent has been created */
 350         void (*notify_create_mad_agent)(struct rvt_dev_info *rdi, int port_idx);
 351
 352         /* Notify driver a mad agent has been removed */
 353         void (*notify_free_mad_agent)(struct rvt_dev_info *rdi, int port_idx);
 354
 355         /* Notify driver to restart rc */
 356         void (*notify_restart_rc)(struct rvt_qp *qp, u32 psn, int wait);
 357 };
 358
 359 struct rvt_dev_info {
 360         struct ib_device ibdev; /* Keep this first. Nothing above here */
 361
 362         /*
 363          * Prior to calling for registration the driver will be responsible for
 364          * allocating space for this structure.
 365          *
 366          * The driver will also be responsible for filling in certain members of
 367          * dparms.props. The driver needs to fill in dparms exactly as it would
 368          * want values reported to a ULP. This will be returned to the caller
 369          * in rdmavt's device. The driver should also therefore refrain from
 370          * modifying this directly after registration with rdmavt.
 371          */
 372
 373         /* Driver specific properties */
 374         struct rvt_driver_params dparms;
 375
 376         /* post send table */
 377         const struct rvt_operation_params *post_parms;
 378
 379         /* Driver specific helper functions */
 380         struct rvt_driver_provided driver_f;
 381
 382         struct rvt_mregion __rcu *dma_mr;
 383         struct rvt_lkey_table lkey_table;
 384
 385         /* Internal use */
 386         int n_pds_allocated;
 387         spinlock_t n_pds_lock; /* Protect pd allocated count */
 388
 389         int n_ahs_allocated;
 390         spinlock_t n_ahs_lock; /* Protect ah allocated count */
 391
 392         u32 n_srqs_allocated;
 393         spinlock_t n_srqs_lock; /* Protect srqs allocated count */
 394
 395         int flags;
 396         struct rvt_ibport **ports;
 397
 398         /* QP */
 399         struct rvt_qp_ibdev *qp_dev;
 400         u32 n_qps_allocated;    /* number of QPs allocated for device */
 401         u32 n_rc_qps;           /* number of RC QPs allocated for device */
 402         u32 busy_jiffies;       /* timeout scaling based on RC QP count */
 403         spinlock_t n_qps_lock;  /* protect qps, rc qps and busy jiffy counts */
 404
 405         /* memory maps */
 406         struct list_head pending_mmaps;
 407         spinlock_t mmap_offset_lock; /* protect mmap_offset */
 408         u32 mmap_offset;
 409         spinlock_t pending_lock; /* protect pending mmap list */
 410
 411         /* CQ */
 412         struct kthread_worker *worker; /* per device cq worker */
 413         u32 n_cqs_allocated;    /* number of CQs allocated for device */
 414         spinlock_t n_cqs_lock; /* protect count of in use cqs */
 415
 416         /* Multicast */
 417         u32 n_mcast_grps_allocated; /* number of mcast groups allocated */
 418         spinlock_t n_mcast_grps_lock;
 419
 420 };
 421
 422 static inline struct rvt_pd *ibpd_to_rvtpd(struct ib_pd *ibpd)
 423 {
 424         return container_of(ibpd, struct rvt_pd, ibpd);
 425 }
 426
 427 static inline struct rvt_ah *ibah_to_rvtah(struct ib_ah *ibah)
 428 {
 429         return container_of(ibah, struct rvt_ah, ibah);
 430 }
 431
 432 static inline struct rvt_dev_info *ib_to_rvt(struct ib_device *ibdev)
 433 {
 434         return  container_of(ibdev, struct rvt_dev_info, ibdev);
 435 }
 436
 437 static inline struct rvt_srq *ibsrq_to_rvtsrq(struct ib_srq *ibsrq)
 438 {
 439         return container_of(ibsrq, struct rvt_srq, ibsrq);
 440 }
 441
 442 static inline struct rvt_qp *ibqp_to_rvtqp(struct ib_qp *ibqp)
 443 {
 444         return container_of(ibqp, struct rvt_qp, ibqp);
 445 }
 446
 447 static inline unsigned rvt_get_npkeys(struct rvt_dev_info *rdi)
 448 {
 449         /*
 450          * All ports have same number of pkeys.
 451          */
 452         return rdi->dparms.npkeys;
 453 }
 454
 455 /*
 456  * Return the max atomic suitable for determining
 457  * the size of the ack ring buffer in a QP.
 458  */
 459 static inline unsigned int rvt_max_atomic(struct rvt_dev_info *rdi)
 460 {
 461         return rdi->dparms.max_rdma_atomic + 1;
 462 }
 463
 464 /*
 465  * Return the indexed PKEY from the port PKEY table.
 466  */
 467 static inline u16 rvt_get_pkey(struct rvt_dev_info *rdi,
 468                                int port_index,
 469                                unsigned index)
 470 {
 471         if (index >= rvt_get_npkeys(rdi))
 472                 return 0;
 473         else
 474                 return rdi->ports[port_index]->pkey_table[index];
 475 }
 476
 477 /**
 478  * rvt_lookup_qpn - return the QP with the given QPN
 479  * @ibp: the ibport
 480  * @qpn: the QP number to look up
 481  *
 482  * The caller must hold the rcu_read_lock(), and keep the lock until
 483  * the returned qp is no longer in use.
 484  */
 485 /* TODO: Remove this and put in rdmavt/qp.h when no longer needed by drivers */
 486 static inline struct rvt_qp *rvt_lookup_qpn(struct rvt_dev_info *rdi,
 487                                             struct rvt_ibport *rvp,
 488                                             u32 qpn) __must_hold(RCU)
 489 {
 490         struct rvt_qp *qp = NULL;
 491
 492         if (unlikely(qpn <= 1)) {
 493                 qp = rcu_dereference(rvp->qp[qpn]);
 494         } else {
 495                 u32 n = hash_32(qpn, rdi->qp_dev->qp_table_bits);
 496
 497                 for (qp = rcu_dereference(rdi->qp_dev->qp_table[n]); qp;
 498                         qp = rcu_dereference(qp->next))
 499                         if (qp->ibqp.qp_num == qpn)
 500                                 break;
 501         }
 502         return qp;
 503 }
 504
 505 /**
 506  * rvt_mod_retry_timer - mod a retry timer
 507  * @qp - the QP
 508  * Modify a potentially already running retry timer
 509  */
 510 static inline void rvt_mod_retry_timer(struct rvt_qp *qp)
 511 {
 512         struct ib_qp *ibqp = &qp->ibqp;
 513         struct rvt_dev_info *rdi = ib_to_rvt(ibqp->device);
 514
 515         lockdep_assert_held(&qp->s_lock);
 516         qp->s_flags |= RVT_S_TIMER;
 517         /* 4.096 usec. * (1 << qp->timeout) */
 518         mod_timer(&qp->s_timer, jiffies + qp->timeout_jiffies +
 519                   rdi->busy_jiffies);
 520 }
 521
 522 struct rvt_dev_info *rvt_alloc_device(size_t size, int nports);
 523 void rvt_dealloc_device(struct rvt_dev_info *rdi);
 524 int rvt_register_device(struct rvt_dev_info *rvd);
 525 void rvt_unregister_device(struct rvt_dev_info *rvd);
 526 int rvt_check_ah(struct ib_device *ibdev, struct rdma_ah_attr *ah_attr);
 527 int rvt_init_port(struct rvt_dev_info *rdi, struct rvt_ibport *port,
 528                   int port_index, u16 *pkey_table);
 529 int rvt_fast_reg_mr(struct rvt_qp *qp, struct ib_mr *ibmr, u32 key,
 530                     int access);
 531 int rvt_invalidate_rkey(struct rvt_qp *qp, u32 rkey);
 532 int rvt_rkey_ok(struct rvt_qp *qp, struct rvt_sge *sge,
 533                 u32 len, u64 vaddr, u32 rkey, int acc);
 534 int rvt_lkey_ok(struct rvt_lkey_table *rkt, struct rvt_pd *pd,
 535                 struct rvt_sge *isge, struct rvt_sge *last_sge,
 536                 struct ib_sge *sge, int acc);
 537 struct rvt_mcast *rvt_mcast_find(struct rvt_ibport *ibp, union ib_gid *mgid,
 538                                  u16 lid);
 539
 540 #endif          /* DEF_RDMA_VT_H */