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/rdmavt_mr.h>
  61 #include <rdma/rdmavt_qp.h>
  62
  63 #define RVT_MAX_PKEY_VALUES 16
  64
  65 struct rvt_ibport {
  66         struct rvt_qp __rcu *qp[2];
  67         struct ib_mad_agent *send_agent;        /* agent for SMI (traps) */
  68         struct rb_root mcast_tree;
  69         spinlock_t lock;                /* protect changes in this struct */
  70
  71         /* non-zero when timer is set */
  72         unsigned long mkey_lease_timeout;
  73         unsigned long trap_timeout;
  74         __be64 gid_prefix;      /* in network order */
  75         __be64 mkey;
  76         u64 tid;
  77         u32 port_cap_flags;
  78         u32 pma_sample_start;
  79         u32 pma_sample_interval;
  80         __be16 pma_counter_select[5];
  81         u16 pma_tag;
  82         u16 mkey_lease_period;
  83         u16 sm_lid;
  84         u8 sm_sl;
  85         u8 mkeyprot;
  86         u8 subnet_timeout;
  87         u8 vl_high_limit;
  88
  89         /*
  90          * Driver is expected to keep these up to date. These
  91          * counters are informational only and not required to be
  92          * completely accurate.
  93          */
  94         u64 n_rc_resends;
  95         u64 n_seq_naks;
  96         u64 n_rdma_seq;
  97         u64 n_rnr_naks;
  98         u64 n_other_naks;
  99         u64 n_loop_pkts;
 100         u64 n_pkt_drops;
 101         u64 n_vl15_dropped;
 102         u64 n_rc_timeouts;
 103         u64 n_dmawait;
 104         u64 n_unaligned;
 105         u64 n_rc_dupreq;
 106         u64 n_rc_seqnak;
 107         u16 pkey_violations;
 108         u16 qkey_violations;
 109         u16 mkey_violations;
 110
 111         /* Hot-path per CPU counters to avoid cacheline trading to update */
 112         u64 z_rc_acks;
 113         u64 z_rc_qacks;
 114         u64 z_rc_delayed_comp;
 115         u64 __percpu *rc_acks;
 116         u64 __percpu *rc_qacks;
 117         u64 __percpu *rc_delayed_comp;
 118
 119         void *priv; /* driver private data */
 120
 121         /*
 122          * The pkey table is allocated and maintained by the driver. Drivers
 123          * need to have access to this before registering with rdmav. However
 124          * rdmavt will need access to it so drivers need to proviee this during
 125          * the attach port API call.
 126          */
 127         u16 *pkey_table;
 128
 129         struct rvt_ah *sm_ah;
 130 };
 131
 132 #define RVT_CQN_MAX 16 /* maximum length of cq name */
 133
 134 /*
 135  * Things that are driver specific, module parameters in hfi1 and qib
 136  */
 137 struct rvt_driver_params {
 138         struct ib_device_attr props;
 139
 140         /*
 141          * Anything driver specific that is not covered by props
 142          * For instance special module parameters. Goes here.
 143          */
 144         unsigned int lkey_table_size;
 145         unsigned int qp_table_size;
 146         int qpn_start;
 147         int qpn_inc;
 148         int qpn_res_start;
 149         int qpn_res_end;
 150         int nports;
 151         int npkeys;
 152         u8 qos_shift;
 153         char cq_name[RVT_CQN_MAX];
 154         int node;
 155         int max_rdma_atomic;
 156         int psn_mask;
 157         int psn_shift;
 158         int psn_modify_mask;
 159         u32 core_cap_flags;
 160         u32 max_mad_size;
 161 };
 162
 163 /* Protection domain */
 164 struct rvt_pd {
 165         struct ib_pd ibpd;
 166         int user;               /* non-zero if created from user space */
 167 };
 168
 169 /* Address handle */
 170 struct rvt_ah {
 171         struct ib_ah ibah;
 172         struct ib_ah_attr attr;
 173         atomic_t refcount;
 174         u8 vl;
 175         u8 log_pmtu;
 176 };
 177
 178 struct rvt_dev_info;
 179 struct rvt_swqe;
 180 struct rvt_driver_provided {
 181         /*
 182          * Which functions are required depends on which verbs rdmavt is
 183          * providing and which verbs the driver is overriding. See
 184          * check_support() for details.
 185          */
 186
 187         /* Passed to ib core registration. Callback to create syfs files */
 188         int (*port_callback)(struct ib_device *, u8, struct kobject *);
 189
 190         /*
 191          * Returns a string to represent the device for which is being
 192          * registered. This is primarily used for error and debug messages on
 193          * the console.
 194          */
 195         const char * (*get_card_name)(struct rvt_dev_info *rdi);
 196
 197         /*
 198          * Returns a pointer to the undelying hardware's PCI device. This is
 199          * used to display information as to what hardware is being referenced
 200          * in an output message
 201          */
 202         struct pci_dev * (*get_pci_dev)(struct rvt_dev_info *rdi);
 203
 204         /*
 205          * Allocate a private queue pair data structure for driver specific
 206          * information which is opaque to rdmavt.
 207          */
 208         void * (*qp_priv_alloc)(struct rvt_dev_info *rdi, struct rvt_qp *qp,
 209                                 gfp_t gfp);
 210
 211         /*
 212          * Free the driver's private qp structure.
 213          */
 214         void (*qp_priv_free)(struct rvt_dev_info *rdi, struct rvt_qp *qp);
 215
 216         /*
 217          * Inform the driver the particular qp in quesiton has been reset so
 218          * that it can clean up anything it needs to.
 219          */
 220         void (*notify_qp_reset)(struct rvt_qp *qp);
 221
 222         /*
 223          * Give the driver a notice that there is send work to do. It is up to
 224          * the driver to generally push the packets out, this just queues the
 225          * work with the driver. There are two variants here. The no_lock
 226          * version requires the s_lock not to be held. The other assumes the
 227          * s_lock is held.
 228          */
 229         void (*schedule_send)(struct rvt_qp *qp);
 230         void (*schedule_send_no_lock)(struct rvt_qp *qp);
 231
 232         /*
 233          * Sometimes rdmavt needs to kick the driver's send progress. That is
 234          * done by this call back.
 235          */
 236         void (*do_send)(struct rvt_qp *qp);
 237
 238         /*
 239          * Get a path mtu from the driver based on qp attributes.
 240          */
 241         int (*get_pmtu_from_attr)(struct rvt_dev_info *rdi, struct rvt_qp *qp,
 242                                   struct ib_qp_attr *attr);
 243
 244         /*
 245          * Notify driver that it needs to flush any outstanding IO requests that
 246          * are waiting on a qp.
 247          */
 248         void (*flush_qp_waiters)(struct rvt_qp *qp);
 249
 250         /*
 251          * Notify driver to stop its queue of sending packets. Nothing else
 252          * should be posted to the queue pair after this has been called.
 253          */
 254         void (*stop_send_queue)(struct rvt_qp *qp);
 255
 256         /*
 257          * Have the drivr drain any in progress operations
 258          */
 259         void (*quiesce_qp)(struct rvt_qp *qp);
 260
 261         /*
 262          * Inform the driver a qp has went to error state.
 263          */
 264         void (*notify_error_qp)(struct rvt_qp *qp);
 265
 266         /*
 267          * Get an MTU for a qp.
 268          */
 269         u32 (*mtu_from_qp)(struct rvt_dev_info *rdi, struct rvt_qp *qp,
 270                            u32 pmtu);
 271         /*
 272          * Convert an mtu to a path mtu
 273          */
 274         int (*mtu_to_path_mtu)(u32 mtu);
 275
 276         /*
 277          * Get the guid of a port in big endian byte order
 278          */
 279         int (*get_guid_be)(struct rvt_dev_info *rdi, struct rvt_ibport *rvp,
 280                            int guid_index, __be64 *guid);
 281
 282         /*
 283          * Query driver for the state of the port.
 284          */
 285         int (*query_port_state)(struct rvt_dev_info *rdi, u8 port_num,
 286                                 struct ib_port_attr *props);
 287
 288         /*
 289          * Tell driver to shutdown a port
 290          */
 291         int (*shut_down_port)(struct rvt_dev_info *rdi, u8 port_num);
 292
 293         /* Tell driver to send a trap for changed  port capabilities */
 294         void (*cap_mask_chg)(struct rvt_dev_info *rdi, u8 port_num);
 295
 296         /*
 297          * The following functions can be safely ignored completely. Any use of
 298          * these is checked for NULL before blindly calling. Rdmavt should also
 299          * be functional if drivers omit these.
 300          */
 301
 302         /* Called to inform the driver that all qps should now be freed. */
 303         unsigned (*free_all_qps)(struct rvt_dev_info *rdi);
 304
 305         /* Driver specific AH validation */
 306         int (*check_ah)(struct ib_device *, struct ib_ah_attr *);
 307
 308         /* Inform the driver a new AH has been created */
 309         void (*notify_new_ah)(struct ib_device *, struct ib_ah_attr *,
 310                               struct rvt_ah *);
 311
 312         /* Let the driver pick the next queue pair number*/
 313         int (*alloc_qpn)(struct rvt_dev_info *rdi, struct rvt_qpn_table *qpt,
 314                          enum ib_qp_type type, u8 port_num, gfp_t gfp);
 315
 316         /* Determine if its safe or allowed to modify the qp */
 317         int (*check_modify_qp)(struct rvt_qp *qp, struct ib_qp_attr *attr,
 318                                int attr_mask, struct ib_udata *udata);
 319
 320         /* Driver specific QP modification/notification-of */
 321         void (*modify_qp)(struct rvt_qp *qp, struct ib_qp_attr *attr,
 322                           int attr_mask, struct ib_udata *udata);
 323
 324         /* Driver specific work request checking */
 325         int (*check_send_wqe)(struct rvt_qp *qp, struct rvt_swqe *wqe);
 326
 327         /* Notify driver a mad agent has been created */
 328         void (*notify_create_mad_agent)(struct rvt_dev_info *rdi, int port_idx);
 329
 330         /* Notify driver a mad agent has been removed */
 331         void (*notify_free_mad_agent)(struct rvt_dev_info *rdi, int port_idx);
 332
 333 };
 334
 335 struct rvt_dev_info {
 336         struct ib_device ibdev; /* Keep this first. Nothing above here */
 337
 338         /*
 339          * Prior to calling for registration the driver will be responsible for
 340          * allocating space for this structure.
 341          *
 342          * The driver will also be responsible for filling in certain members of
 343          * dparms.props. The driver needs to fill in dparms exactly as it would
 344          * want values reported to a ULP. This will be returned to the caller
 345          * in rdmavt's device. The driver should also therefore refrain from
 346          * modifying this directly after registration with rdmavt.
 347          */
 348
 349         /* Driver specific properties */
 350         struct rvt_driver_params dparms;
 351
 352         struct rvt_mregion __rcu *dma_mr;
 353         struct rvt_lkey_table lkey_table;
 354
 355         /* Driver specific helper functions */
 356         struct rvt_driver_provided driver_f;
 357
 358         /* Internal use */
 359         int n_pds_allocated;
 360         spinlock_t n_pds_lock; /* Protect pd allocated count */
 361
 362         int n_ahs_allocated;
 363         spinlock_t n_ahs_lock; /* Protect ah allocated count */
 364
 365         u32 n_srqs_allocated;
 366         spinlock_t n_srqs_lock; /* Protect srqs allocated count */
 367
 368         int flags;
 369         struct rvt_ibport **ports;
 370
 371         /* QP */
 372         struct rvt_qp_ibdev *qp_dev;
 373         u32 n_qps_allocated;    /* number of QPs allocated for device */
 374         u32 n_rc_qps;           /* number of RC QPs allocated for device */
 375         u32 busy_jiffies;       /* timeout scaling based on RC QP count */
 376         spinlock_t n_qps_lock;  /* protect qps, rc qps and busy jiffy counts */
 377
 378         /* memory maps */
 379         struct list_head pending_mmaps;
 380         spinlock_t mmap_offset_lock; /* protect mmap_offset */
 381         u32 mmap_offset;
 382         spinlock_t pending_lock; /* protect pending mmap list */
 383
 384         /* CQ */
 385         struct kthread_worker *worker; /* per device cq worker */
 386         u32 n_cqs_allocated;    /* number of CQs allocated for device */
 387         spinlock_t n_cqs_lock; /* protect count of in use cqs */
 388
 389         /* Multicast */
 390         u32 n_mcast_grps_allocated; /* number of mcast groups allocated */
 391         spinlock_t n_mcast_grps_lock;
 392
 393 };
 394
 395 static inline struct rvt_pd *ibpd_to_rvtpd(struct ib_pd *ibpd)
 396 {
 397         return container_of(ibpd, struct rvt_pd, ibpd);
 398 }
 399
 400 static inline struct rvt_ah *ibah_to_rvtah(struct ib_ah *ibah)
 401 {
 402         return container_of(ibah, struct rvt_ah, ibah);
 403 }
 404
 405 static inline struct rvt_dev_info *ib_to_rvt(struct ib_device *ibdev)
 406 {
 407         return  container_of(ibdev, struct rvt_dev_info, ibdev);
 408 }
 409
 410 static inline struct rvt_srq *ibsrq_to_rvtsrq(struct ib_srq *ibsrq)
 411 {
 412         return container_of(ibsrq, struct rvt_srq, ibsrq);
 413 }
 414
 415 static inline struct rvt_qp *ibqp_to_rvtqp(struct ib_qp *ibqp)
 416 {
 417         return container_of(ibqp, struct rvt_qp, ibqp);
 418 }
 419
 420 static inline unsigned rvt_get_npkeys(struct rvt_dev_info *rdi)
 421 {
 422         /*
 423          * All ports have same number of pkeys.
 424          */
 425         return rdi->dparms.npkeys;
 426 }
 427
 428 /*
 429  * Return the indexed PKEY from the port PKEY table.
 430  */
 431 static inline u16 rvt_get_pkey(struct rvt_dev_info *rdi,
 432                                int port_index,
 433                                unsigned index)
 434 {
 435         if (index >= rvt_get_npkeys(rdi))
 436                 return 0;
 437         else
 438                 return rdi->ports[port_index]->pkey_table[index];
 439 }
 440
 441 /**
 442  * rvt_lookup_qpn - return the QP with the given QPN
 443  * @ibp: the ibport
 444  * @qpn: the QP number to look up
 445  *
 446  * The caller must hold the rcu_read_lock(), and keep the lock until
 447  * the returned qp is no longer in use.
 448  */
 449 /* TODO: Remove this and put in rdmavt/qp.h when no longer needed by drivers */
 450 static inline struct rvt_qp *rvt_lookup_qpn(struct rvt_dev_info *rdi,
 451                                             struct rvt_ibport *rvp,
 452                                             u32 qpn) __must_hold(RCU)
 453 {
 454         struct rvt_qp *qp = NULL;
 455
 456         if (unlikely(qpn <= 1)) {
 457                 qp = rcu_dereference(rvp->qp[qpn]);
 458         } else {
 459                 u32 n = hash_32(qpn, rdi->qp_dev->qp_table_bits);
 460
 461                 for (qp = rcu_dereference(rdi->qp_dev->qp_table[n]); qp;
 462                         qp = rcu_dereference(qp->next))
 463                         if (qp->ibqp.qp_num == qpn)
 464                                 break;
 465         }
 466         return qp;
 467 }
 468
 469 struct rvt_dev_info *rvt_alloc_device(size_t size, int nports);
 470 int rvt_register_device(struct rvt_dev_info *rvd);
 471 void rvt_unregister_device(struct rvt_dev_info *rvd);
 472 int rvt_check_ah(struct ib_device *ibdev, struct ib_ah_attr *ah_attr);
 473 int rvt_init_port(struct rvt_dev_info *rdi, struct rvt_ibport *port,
 474                   int port_index, u16 *pkey_table);
 475 int rvt_rkey_ok(struct rvt_qp *qp, struct rvt_sge *sge,
 476                 u32 len, u64 vaddr, u32 rkey, int acc);
 477 int rvt_lkey_ok(struct rvt_lkey_table *rkt, struct rvt_pd *pd,
 478                 struct rvt_sge *isge, struct ib_sge *sge, int acc);
 479 struct rvt_mcast *rvt_mcast_find(struct rvt_ibport *ibp, union ib_gid *mgid);
 480
 481 #endif          /* DEF_RDMA_VT_H */