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         char cq_name[RVT_CQN_MAX];
 153         int node;
 154         int psn_mask;
 155         int psn_shift;
 156         int psn_modify_mask;
 157         u32 core_cap_flags;
 158         u32 max_mad_size;
 159         u8 qos_shift;
 160         u8 max_rdma_atomic;
 161         u8 reserved_operations;
 162 };
 163
 164 /* Protection domain */
 165 struct rvt_pd {
 166         struct ib_pd ibpd;
 167         bool user;
 168 };
 169
 170 /* Address handle */
 171 struct rvt_ah {
 172         struct ib_ah ibah;
 173         struct ib_ah_attr attr;
 174         atomic_t refcount;
 175         u8 vl;
 176         u8 log_pmtu;
 177 };
 178
 179 struct rvt_dev_info;
 180 struct rvt_swqe;
 181 struct rvt_driver_provided {
 182         /*
 183          * Which functions are required depends on which verbs rdmavt is
 184          * providing and which verbs the driver is overriding. See
 185          * check_support() for details.
 186          */
 187
 188         /* hot path calldowns in a single cacheline */
 189
 190         /*
 191          * Give the driver a notice that there is send work to do. It is up to
 192          * the driver to generally push the packets out, this just queues the
 193          * work with the driver. There are two variants here. The no_lock
 194          * version requires the s_lock not to be held. The other assumes the
 195          * s_lock is held.
 196          */
 197         void (*schedule_send)(struct rvt_qp *qp);
 198         void (*schedule_send_no_lock)(struct rvt_qp *qp);
 199
 200         /* Driver specific work request checking */
 201         int (*check_send_wqe)(struct rvt_qp *qp, struct rvt_swqe *wqe);
 202
 203         /*
 204          * Sometimes rdmavt needs to kick the driver's send progress. That is
 205          * done by this call back.
 206          */
 207         void (*do_send)(struct rvt_qp *qp);
 208
 209         /* Passed to ib core registration. Callback to create syfs files */
 210         int (*port_callback)(struct ib_device *, u8, struct kobject *);
 211
 212         /*
 213          * Returns a string to represent the device for which is being
 214          * registered. This is primarily used for error and debug messages on
 215          * the console.
 216          */
 217         const char * (*get_card_name)(struct rvt_dev_info *rdi);
 218
 219         /*
 220          * Returns a pointer to the undelying hardware's PCI device. This is
 221          * used to display information as to what hardware is being referenced
 222          * in an output message
 223          */
 224         struct pci_dev * (*get_pci_dev)(struct rvt_dev_info *rdi);
 225
 226         /*
 227          * Allocate a private queue pair data structure for driver specific
 228          * information which is opaque to rdmavt.  Errors are returned via
 229          * ERR_PTR(err).  The driver is free to return NULL or a valid
 230          * pointer.
 231          */
 232         void * (*qp_priv_alloc)(struct rvt_dev_info *rdi, struct rvt_qp *qp,
 233                                 gfp_t gfp);
 234
 235         /*
 236          * Free the driver's private qp structure.
 237          */
 238         void (*qp_priv_free)(struct rvt_dev_info *rdi, struct rvt_qp *qp);
 239
 240         /*
 241          * Inform the driver the particular qp in quesiton has been reset so
 242          * that it can clean up anything it needs to.
 243          */
 244         void (*notify_qp_reset)(struct rvt_qp *qp);
 245
 246         /*
 247          * Get a path mtu from the driver based on qp attributes.
 248          */
 249         int (*get_pmtu_from_attr)(struct rvt_dev_info *rdi, struct rvt_qp *qp,
 250                                   struct ib_qp_attr *attr);
 251
 252         /*
 253          * Notify driver that it needs to flush any outstanding IO requests that
 254          * are waiting on a qp.
 255          */
 256         void (*flush_qp_waiters)(struct rvt_qp *qp);
 257
 258         /*
 259          * Notify driver to stop its queue of sending packets. Nothing else
 260          * should be posted to the queue pair after this has been called.
 261          */
 262         void (*stop_send_queue)(struct rvt_qp *qp);
 263
 264         /*
 265          * Have the drivr drain any in progress operations
 266          */
 267         void (*quiesce_qp)(struct rvt_qp *qp);
 268
 269         /*
 270          * Inform the driver a qp has went to error state.
 271          */
 272         void (*notify_error_qp)(struct rvt_qp *qp);
 273
 274         /*
 275          * Get an MTU for a qp.
 276          */
 277         u32 (*mtu_from_qp)(struct rvt_dev_info *rdi, struct rvt_qp *qp,
 278                            u32 pmtu);
 279         /*
 280          * Convert an mtu to a path mtu
 281          */
 282         int (*mtu_to_path_mtu)(u32 mtu);
 283
 284         /*
 285          * Get the guid of a port in big endian byte order
 286          */
 287         int (*get_guid_be)(struct rvt_dev_info *rdi, struct rvt_ibport *rvp,
 288                            int guid_index, __be64 *guid);
 289
 290         /*
 291          * Query driver for the state of the port.
 292          */
 293         int (*query_port_state)(struct rvt_dev_info *rdi, u8 port_num,
 294                                 struct ib_port_attr *props);
 295
 296         /*
 297          * Tell driver to shutdown a port
 298          */
 299         int (*shut_down_port)(struct rvt_dev_info *rdi, u8 port_num);
 300
 301         /* Tell driver to send a trap for changed  port capabilities */
 302         void (*cap_mask_chg)(struct rvt_dev_info *rdi, u8 port_num);
 303
 304         /*
 305          * The following functions can be safely ignored completely. Any use of
 306          * these is checked for NULL before blindly calling. Rdmavt should also
 307          * be functional if drivers omit these.
 308          */
 309
 310         /* Called to inform the driver that all qps should now be freed. */
 311         unsigned (*free_all_qps)(struct rvt_dev_info *rdi);
 312
 313         /* Driver specific AH validation */
 314         int (*check_ah)(struct ib_device *, struct ib_ah_attr *);
 315
 316         /* Inform the driver a new AH has been created */
 317         void (*notify_new_ah)(struct ib_device *, struct ib_ah_attr *,
 318                               struct rvt_ah *);
 319
 320         /* Let the driver pick the next queue pair number*/
 321         int (*alloc_qpn)(struct rvt_dev_info *rdi, struct rvt_qpn_table *qpt,
 322                          enum ib_qp_type type, u8 port_num, gfp_t gfp);
 323
 324         /* Determine if its safe or allowed to modify the qp */
 325         int (*check_modify_qp)(struct rvt_qp *qp, struct ib_qp_attr *attr,
 326                                int attr_mask, struct ib_udata *udata);
 327
 328         /* Driver specific QP modification/notification-of */
 329         void (*modify_qp)(struct rvt_qp *qp, struct ib_qp_attr *attr,
 330                           int attr_mask, struct ib_udata *udata);
 331
 332         /* Notify driver a mad agent has been created */
 333         void (*notify_create_mad_agent)(struct rvt_dev_info *rdi, int port_idx);
 334
 335         /* Notify driver a mad agent has been removed */
 336         void (*notify_free_mad_agent)(struct rvt_dev_info *rdi, int port_idx);
 337
 338         /* Notify driver to restart rc */
 339         void (*notify_restart_rc)(struct rvt_qp *qp, u32 psn, int wait);
 340 };
 341
 342 struct rvt_dev_info {
 343         struct ib_device ibdev; /* Keep this first. Nothing above here */
 344
 345         /*
 346          * Prior to calling for registration the driver will be responsible for
 347          * allocating space for this structure.
 348          *
 349          * The driver will also be responsible for filling in certain members of
 350          * dparms.props. The driver needs to fill in dparms exactly as it would
 351          * want values reported to a ULP. This will be returned to the caller
 352          * in rdmavt's device. The driver should also therefore refrain from
 353          * modifying this directly after registration with rdmavt.
 354          */
 355
 356         /* Driver specific properties */
 357         struct rvt_driver_params dparms;
 358
 359         /* post send table */
 360         const struct rvt_operation_params *post_parms;
 361
 362         /* Driver specific helper functions */
 363         struct rvt_driver_provided driver_f;
 364
 365         struct rvt_mregion __rcu *dma_mr;
 366         struct rvt_lkey_table lkey_table;
 367
 368         /* Internal use */
 369         int n_pds_allocated;
 370         spinlock_t n_pds_lock; /* Protect pd allocated count */
 371
 372         int n_ahs_allocated;
 373         spinlock_t n_ahs_lock; /* Protect ah allocated count */
 374
 375         u32 n_srqs_allocated;
 376         spinlock_t n_srqs_lock; /* Protect srqs allocated count */
 377
 378         int flags;
 379         struct rvt_ibport **ports;
 380
 381         /* QP */
 382         struct rvt_qp_ibdev *qp_dev;
 383         u32 n_qps_allocated;    /* number of QPs allocated for device */
 384         u32 n_rc_qps;           /* number of RC QPs allocated for device */
 385         u32 busy_jiffies;       /* timeout scaling based on RC QP count */
 386         spinlock_t n_qps_lock;  /* protect qps, rc qps and busy jiffy counts */
 387
 388         /* memory maps */
 389         struct list_head pending_mmaps;
 390         spinlock_t mmap_offset_lock; /* protect mmap_offset */
 391         u32 mmap_offset;
 392         spinlock_t pending_lock; /* protect pending mmap list */
 393
 394         /* CQ */
 395         struct kthread_worker *worker; /* per device cq worker */
 396         u32 n_cqs_allocated;    /* number of CQs allocated for device */
 397         spinlock_t n_cqs_lock; /* protect count of in use cqs */
 398
 399         /* Multicast */
 400         u32 n_mcast_grps_allocated; /* number of mcast groups allocated */
 401         spinlock_t n_mcast_grps_lock;
 402
 403 };
 404
 405 static inline struct rvt_pd *ibpd_to_rvtpd(struct ib_pd *ibpd)
 406 {
 407         return container_of(ibpd, struct rvt_pd, ibpd);
 408 }
 409
 410 static inline struct rvt_ah *ibah_to_rvtah(struct ib_ah *ibah)
 411 {
 412         return container_of(ibah, struct rvt_ah, ibah);
 413 }
 414
 415 static inline struct rvt_dev_info *ib_to_rvt(struct ib_device *ibdev)
 416 {
 417         return  container_of(ibdev, struct rvt_dev_info, ibdev);
 418 }
 419
 420 static inline struct rvt_srq *ibsrq_to_rvtsrq(struct ib_srq *ibsrq)
 421 {
 422         return container_of(ibsrq, struct rvt_srq, ibsrq);
 423 }
 424
 425 static inline struct rvt_qp *ibqp_to_rvtqp(struct ib_qp *ibqp)
 426 {
 427         return container_of(ibqp, struct rvt_qp, ibqp);
 428 }
 429
 430 static inline unsigned rvt_get_npkeys(struct rvt_dev_info *rdi)
 431 {
 432         /*
 433          * All ports have same number of pkeys.
 434          */
 435         return rdi->dparms.npkeys;
 436 }
 437
 438 /*
 439  * Return the max atomic suitable for determining
 440  * the size of the ack ring buffer in a QP.
 441  */
 442 static inline unsigned int rvt_max_atomic(struct rvt_dev_info *rdi)
 443 {
 444         return rdi->dparms.max_rdma_atomic + 1;
 445 }
 446
 447 /*
 448  * Return the indexed PKEY from the port PKEY table.
 449  */
 450 static inline u16 rvt_get_pkey(struct rvt_dev_info *rdi,
 451                                int port_index,
 452                                unsigned index)
 453 {
 454         if (index >= rvt_get_npkeys(rdi))
 455                 return 0;
 456         else
 457                 return rdi->ports[port_index]->pkey_table[index];
 458 }
 459
 460 /**
 461  * rvt_lookup_qpn - return the QP with the given QPN
 462  * @ibp: the ibport
 463  * @qpn: the QP number to look up
 464  *
 465  * The caller must hold the rcu_read_lock(), and keep the lock until
 466  * the returned qp is no longer in use.
 467  */
 468 /* TODO: Remove this and put in rdmavt/qp.h when no longer needed by drivers */
 469 static inline struct rvt_qp *rvt_lookup_qpn(struct rvt_dev_info *rdi,
 470                                             struct rvt_ibport *rvp,
 471                                             u32 qpn) __must_hold(RCU)
 472 {
 473         struct rvt_qp *qp = NULL;
 474
 475         if (unlikely(qpn <= 1)) {
 476                 qp = rcu_dereference(rvp->qp[qpn]);
 477         } else {
 478                 u32 n = hash_32(qpn, rdi->qp_dev->qp_table_bits);
 479
 480                 for (qp = rcu_dereference(rdi->qp_dev->qp_table[n]); qp;
 481                         qp = rcu_dereference(qp->next))
 482                         if (qp->ibqp.qp_num == qpn)
 483                                 break;
 484         }
 485         return qp;
 486 }
 487
 488 /**
 489  * rvt_mod_retry_timer - mod a retry timer
 490  * @qp - the QP
 491  * Modify a potentially already running retry timer
 492  */
 493 static inline void rvt_mod_retry_timer(struct rvt_qp *qp)
 494 {
 495         struct ib_qp *ibqp = &qp->ibqp;
 496         struct rvt_dev_info *rdi = ib_to_rvt(ibqp->device);
 497
 498         lockdep_assert_held(&qp->s_lock);
 499         qp->s_flags |= RVT_S_TIMER;
 500         /* 4.096 usec. * (1 << qp->timeout) */
 501         mod_timer(&qp->s_timer, jiffies + qp->timeout_jiffies +
 502                   rdi->busy_jiffies);
 503 }
 504
 505 struct rvt_dev_info *rvt_alloc_device(size_t size, int nports);
 506 void rvt_dealloc_device(struct rvt_dev_info *rdi);
 507 int rvt_register_device(struct rvt_dev_info *rvd);
 508 void rvt_unregister_device(struct rvt_dev_info *rvd);
 509 int rvt_check_ah(struct ib_device *ibdev, struct ib_ah_attr *ah_attr);
 510 int rvt_init_port(struct rvt_dev_info *rdi, struct rvt_ibport *port,
 511                   int port_index, u16 *pkey_table);
 512 int rvt_fast_reg_mr(struct rvt_qp *qp, struct ib_mr *ibmr, u32 key,
 513                     int access);
 514 int rvt_invalidate_rkey(struct rvt_qp *qp, u32 rkey);
 515 int rvt_rkey_ok(struct rvt_qp *qp, struct rvt_sge *sge,
 516                 u32 len, u64 vaddr, u32 rkey, int acc);
 517 int rvt_lkey_ok(struct rvt_lkey_table *rkt, struct rvt_pd *pd,
 518                 struct rvt_sge *isge, struct ib_sge *sge, int acc);
 519 struct rvt_mcast *rvt_mcast_find(struct rvt_ibport *ibp, union ib_gid *mgid);
 520
 521 #endif          /* DEF_RDMA_VT_H */