include/rdma/rdma_vt.h

   1 #ifndef DEF_RDMA_VT_H
   2 #define DEF_RDMA_VT_H
   3
   4 /*
   5  * Copyright(c) 2016 - 2019 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
  63 #define RVT_MAX_PKEY_VALUES 16
  64
  65 #define RVT_MAX_TRAP_LEN 100 /* Limit pending trap list */
  66 #define RVT_MAX_TRAP_LISTS 5 /*((IB_NOTICE_TYPE_INFO & 0x0F) + 1)*/
  67 #define RVT_TRAP_TIMEOUT 4096 /* 4.096 usec */
  68
  69 struct trap_list {
  70         u32 list_len;
  71         struct list_head list;
  72 };
  73
  74 struct rvt_qp;
  75 struct rvt_qpn_table;
  76 struct rvt_ibport {
  77         struct rvt_qp __rcu *qp[2];
  78         struct ib_mad_agent *send_agent;        /* agent for SMI (traps) */
  79         struct rb_root mcast_tree;
  80         spinlock_t lock;                /* protect changes in this struct */
  81
  82         /* non-zero when timer is set */
  83         unsigned long mkey_lease_timeout;
  84         unsigned long trap_timeout;
  85         __be64 gid_prefix;      /* in network order */
  86         __be64 mkey;
  87         u64 tid;
  88         u32 port_cap_flags;
  89         u16 port_cap3_flags;
  90         u32 pma_sample_start;
  91         u32 pma_sample_interval;
  92         __be16 pma_counter_select[5];
  93         u16 pma_tag;
  94         u16 mkey_lease_period;
  95         u32 sm_lid;
  96         u8 sm_sl;
  97         u8 mkeyprot;
  98         u8 subnet_timeout;
  99         u8 vl_high_limit;
 100
 101         /*
 102          * Driver is expected to keep these up to date. These
 103          * counters are informational only and not required to be
 104          * completely accurate.
 105          */
 106         u64 n_rc_resends;
 107         u64 n_seq_naks;
 108         u64 n_rdma_seq;
 109         u64 n_rnr_naks;
 110         u64 n_other_naks;
 111         u64 n_loop_pkts;
 112         u64 n_pkt_drops;
 113         u64 n_vl15_dropped;
 114         u64 n_rc_timeouts;
 115         u64 n_dmawait;
 116         u64 n_unaligned;
 117         u64 n_rc_dupreq;
 118         u64 n_rc_seqnak;
 119         u64 n_rc_crwaits;
 120         u16 pkey_violations;
 121         u16 qkey_violations;
 122         u16 mkey_violations;
 123
 124         /* Hot-path per CPU counters to avoid cacheline trading to update */
 125         u64 z_rc_acks;
 126         u64 z_rc_qacks;
 127         u64 z_rc_delayed_comp;
 128         u64 __percpu *rc_acks;
 129         u64 __percpu *rc_qacks;
 130         u64 __percpu *rc_delayed_comp;
 131
 132         void *priv; /* driver private data */
 133
 134         /*
 135          * The pkey table is allocated and maintained by the driver. Drivers
 136          * need to have access to this before registering with rdmav. However
 137          * rdmavt will need access to it so drivers need to proviee this during
 138          * the attach port API call.
 139          */
 140         u16 *pkey_table;
 141
 142         struct rvt_ah *sm_ah;
 143
 144         /*
 145          * Keep a list of traps that have not been repressed.  They will be
 146          * resent based on trap_timer.
 147          */
 148         struct trap_list trap_lists[RVT_MAX_TRAP_LISTS];
 149         struct timer_list trap_timer;
 150 };
 151
 152 #define RVT_CQN_MAX 16 /* maximum length of cq name */
 153
 154 #define RVT_SGE_COPY_MEMCPY     0
 155 #define RVT_SGE_COPY_CACHELESS  1
 156 #define RVT_SGE_COPY_ADAPTIVE   2
 157
 158 /*
 159  * Things that are driver specific, module parameters in hfi1 and qib
 160  */
 161 struct rvt_driver_params {
 162         struct ib_device_attr props;
 163
 164         /*
 165          * Anything driver specific that is not covered by props
 166          * For instance special module parameters. Goes here.
 167          */
 168         unsigned int lkey_table_size;
 169         unsigned int qp_table_size;
 170         unsigned int sge_copy_mode;
 171         unsigned int wss_threshold;
 172         unsigned int wss_clean_period;
 173         int qpn_start;
 174         int qpn_inc;
 175         int qpn_res_start;
 176         int qpn_res_end;
 177         int nports;
 178         int npkeys;
 179         int node;
 180         int psn_mask;
 181         int psn_shift;
 182         int psn_modify_mask;
 183         u32 core_cap_flags;
 184         u32 max_mad_size;
 185         u8 qos_shift;
 186         u8 max_rdma_atomic;
 187         u8 extra_rdma_atomic;
 188         u8 reserved_operations;
 189 };
 190
 191 /* User context */
 192 struct rvt_ucontext {
 193         struct ib_ucontext ibucontext;
 194 };
 195
 196 /* Protection domain */
 197 struct rvt_pd {
 198         struct ib_pd ibpd;
 199         bool user;
 200 };
 201
 202 /* Address handle */
 203 struct rvt_ah {
 204         struct ib_ah ibah;
 205         struct rdma_ah_attr attr;
 206         u8 vl;
 207         u8 log_pmtu;
 208 };
 209
 210 /*
 211  * This structure is used by rvt_mmap() to validate an offset
 212  * when an mmap() request is made.  The vm_area_struct then uses
 213  * this as its vm_private_data.
 214  */
 215 struct rvt_mmap_info {
 216         struct list_head pending_mmaps;
 217         struct ib_ucontext *context;
 218         void *obj;
 219         __u64 offset;
 220         struct kref ref;
 221         u32 size;
 222 };
 223
 224 /* memory working set size */
 225 struct rvt_wss {
 226         unsigned long *entries;
 227         atomic_t total_count;
 228         atomic_t clean_counter;
 229         atomic_t clean_entry;
 230
 231         int threshold;
 232         int num_entries;
 233         long pages_mask;
 234         unsigned int clean_period;
 235 };
 236
 237 struct rvt_dev_info;
 238 struct rvt_swqe;
 239 struct rvt_driver_provided {
 240         /*
 241          * Which functions are required depends on which verbs rdmavt is
 242          * providing and which verbs the driver is overriding. See
 243          * check_support() for details.
 244          */
 245
 246         /* hot path calldowns in a single cacheline */
 247
 248         /*
 249          * Give the driver a notice that there is send work to do. It is up to
 250          * the driver to generally push the packets out, this just queues the
 251          * work with the driver. There are two variants here. The no_lock
 252          * version requires the s_lock not to be held. The other assumes the
 253          * s_lock is held.
 254          */
 255         bool (*schedule_send)(struct rvt_qp *qp);
 256         bool (*schedule_send_no_lock)(struct rvt_qp *qp);
 257
 258         /*
 259          * Driver specific work request setup and checking.
 260          * This function is allowed to perform any setup, checks, or
 261          * adjustments required to the SWQE in order to be usable by
 262          * underlying protocols. This includes private data structure
 263          * allocations.
 264          */
 265         int (*setup_wqe)(struct rvt_qp *qp, struct rvt_swqe *wqe,
 266                          bool *call_send);
 267
 268         /*
 269          * Sometimes rdmavt needs to kick the driver's send progress. That is
 270          * done by this call back.
 271          */
 272         void (*do_send)(struct rvt_qp *qp);
 273
 274         /*
 275          * Returns a pointer to the undelying hardware's PCI device. This is
 276          * used to display information as to what hardware is being referenced
 277          * in an output message
 278          */
 279         struct pci_dev * (*get_pci_dev)(struct rvt_dev_info *rdi);
 280
 281         /*
 282          * Allocate a private queue pair data structure for driver specific
 283          * information which is opaque to rdmavt.  Errors are returned via
 284          * ERR_PTR(err).  The driver is free to return NULL or a valid
 285          * pointer.
 286          */
 287         void * (*qp_priv_alloc)(struct rvt_dev_info *rdi, struct rvt_qp *qp);
 288
 289         /*
 290          * Init a struture allocated with qp_priv_alloc(). This should be
 291          * called after all qp fields have been initialized in rdmavt.
 292          */
 293         int (*qp_priv_init)(struct rvt_dev_info *rdi, struct rvt_qp *qp,
 294                             struct ib_qp_init_attr *init_attr);
 295
 296         /*
 297          * Free the driver's private qp structure.
 298          */
 299         void (*qp_priv_free)(struct rvt_dev_info *rdi, struct rvt_qp *qp);
 300
 301         /*
 302          * Inform the driver the particular qp in quesiton has been reset so
 303          * that it can clean up anything it needs to.
 304          */
 305         void (*notify_qp_reset)(struct rvt_qp *qp);
 306
 307         /*
 308          * Get a path mtu from the driver based on qp attributes.
 309          */
 310         int (*get_pmtu_from_attr)(struct rvt_dev_info *rdi, struct rvt_qp *qp,
 311                                   struct ib_qp_attr *attr);
 312
 313         /*
 314          * Notify driver that it needs to flush any outstanding IO requests that
 315          * are waiting on a qp.
 316          */
 317         void (*flush_qp_waiters)(struct rvt_qp *qp);
 318
 319         /*
 320          * Notify driver to stop its queue of sending packets. Nothing else
 321          * should be posted to the queue pair after this has been called.
 322          */
 323         void (*stop_send_queue)(struct rvt_qp *qp);
 324
 325         /*
 326          * Have the drivr drain any in progress operations
 327          */
 328         void (*quiesce_qp)(struct rvt_qp *qp);
 329
 330         /*
 331          * Inform the driver a qp has went to error state.
 332          */
 333         void (*notify_error_qp)(struct rvt_qp *qp);
 334
 335         /*
 336          * Get an MTU for a qp.
 337          */
 338         u32 (*mtu_from_qp)(struct rvt_dev_info *rdi, struct rvt_qp *qp,
 339                            u32 pmtu);
 340         /*
 341          * Convert an mtu to a path mtu
 342          */
 343         int (*mtu_to_path_mtu)(u32 mtu);
 344
 345         /*
 346          * Get the guid of a port in big endian byte order
 347          */
 348         int (*get_guid_be)(struct rvt_dev_info *rdi, struct rvt_ibport *rvp,
 349                            int guid_index, __be64 *guid);
 350
 351         /*
 352          * Query driver for the state of the port.
 353          */
 354         int (*query_port_state)(struct rvt_dev_info *rdi, u8 port_num,
 355                                 struct ib_port_attr *props);
 356
 357         /*
 358          * Tell driver to shutdown a port
 359          */
 360         int (*shut_down_port)(struct rvt_dev_info *rdi, u8 port_num);
 361
 362         /* Tell driver to send a trap for changed  port capabilities */
 363         void (*cap_mask_chg)(struct rvt_dev_info *rdi, u8 port_num);
 364
 365         /*
 366          * The following functions can be safely ignored completely. Any use of
 367          * these is checked for NULL before blindly calling. Rdmavt should also
 368          * be functional if drivers omit these.
 369          */
 370
 371         /* Called to inform the driver that all qps should now be freed. */
 372         unsigned (*free_all_qps)(struct rvt_dev_info *rdi);
 373
 374         /* Driver specific AH validation */
 375         int (*check_ah)(struct ib_device *, struct rdma_ah_attr *);
 376
 377         /* Inform the driver a new AH has been created */
 378         void (*notify_new_ah)(struct ib_device *, struct rdma_ah_attr *,
 379                               struct rvt_ah *);
 380
 381         /* Let the driver pick the next queue pair number*/
 382         int (*alloc_qpn)(struct rvt_dev_info *rdi, struct rvt_qpn_table *qpt,
 383                          enum ib_qp_type type, u8 port_num);
 384
 385         /* Determine if its safe or allowed to modify the qp */
 386         int (*check_modify_qp)(struct rvt_qp *qp, struct ib_qp_attr *attr,
 387                                int attr_mask, struct ib_udata *udata);
 388
 389         /* Driver specific QP modification/notification-of */
 390         void (*modify_qp)(struct rvt_qp *qp, struct ib_qp_attr *attr,
 391                           int attr_mask, struct ib_udata *udata);
 392
 393         /* Notify driver a mad agent has been created */
 394         void (*notify_create_mad_agent)(struct rvt_dev_info *rdi, int port_idx);
 395
 396         /* Notify driver a mad agent has been removed */
 397         void (*notify_free_mad_agent)(struct rvt_dev_info *rdi, int port_idx);
 398
 399         /* Notify driver to restart rc */
 400         void (*notify_restart_rc)(struct rvt_qp *qp, u32 psn, int wait);
 401
 402         /* Get and return CPU to pin CQ processing thread */
 403         int (*comp_vect_cpu_lookup)(struct rvt_dev_info *rdi, int comp_vect);
 404 };
 405
 406 struct rvt_dev_info {
 407         struct ib_device ibdev; /* Keep this first. Nothing above here */
 408
 409         /*
 410          * Prior to calling for registration the driver will be responsible for
 411          * allocating space for this structure.
 412          *
 413          * The driver will also be responsible for filling in certain members of
 414          * dparms.props. The driver needs to fill in dparms exactly as it would
 415          * want values reported to a ULP. This will be returned to the caller
 416          * in rdmavt's device. The driver should also therefore refrain from
 417          * modifying this directly after registration with rdmavt.
 418          */
 419
 420         /* Driver specific properties */
 421         struct rvt_driver_params dparms;
 422
 423         /* post send table */
 424         const struct rvt_operation_params *post_parms;
 425
 426         /* opcode translation table */
 427         const enum ib_wc_opcode *wc_opcode;
 428
 429         /* Driver specific helper functions */
 430         struct rvt_driver_provided driver_f;
 431
 432         struct rvt_mregion __rcu *dma_mr;
 433         struct rvt_lkey_table lkey_table;
 434
 435         /* Internal use */
 436         int n_pds_allocated;
 437         spinlock_t n_pds_lock; /* Protect pd allocated count */
 438
 439         int n_ahs_allocated;
 440         spinlock_t n_ahs_lock; /* Protect ah allocated count */
 441
 442         u32 n_srqs_allocated;
 443         spinlock_t n_srqs_lock; /* Protect srqs allocated count */
 444
 445         int flags;
 446         struct rvt_ibport **ports;
 447
 448         /* QP */
 449         struct rvt_qp_ibdev *qp_dev;
 450         u32 n_qps_allocated;    /* number of QPs allocated for device */
 451         u32 n_rc_qps;           /* number of RC QPs allocated for device */
 452         u32 busy_jiffies;       /* timeout scaling based on RC QP count */
 453         spinlock_t n_qps_lock;  /* protect qps, rc qps and busy jiffy counts */
 454
 455         /* memory maps */
 456         struct list_head pending_mmaps;
 457         spinlock_t mmap_offset_lock; /* protect mmap_offset */
 458         u32 mmap_offset;
 459         spinlock_t pending_lock; /* protect pending mmap list */
 460
 461         /* CQ */
 462         u32 n_cqs_allocated;    /* number of CQs allocated for device */
 463         spinlock_t n_cqs_lock; /* protect count of in use cqs */
 464
 465         /* Multicast */
 466         u32 n_mcast_grps_allocated; /* number of mcast groups allocated */
 467         spinlock_t n_mcast_grps_lock;
 468
 469         /* Memory Working Set Size */
 470         struct rvt_wss *wss;
 471 };
 472
 473 /**
 474  * rvt_set_ibdev_name - Craft an IB device name from client info
 475  * @rdi: pointer to the client rvt_dev_info structure
 476  * @name: client specific name
 477  * @unit: client specific unit number.
 478  */
 479 static inline void rvt_set_ibdev_name(struct rvt_dev_info *rdi,
 480                                       const char *fmt, const char *name,
 481                                       const int unit)
 482 {
 483         /*
 484          * FIXME: rvt and its users want to touch the ibdev before
 485          * registration and have things like the name work. We don't have the
 486          * infrastructure in the core to support this directly today, hack it
 487          * to work by setting the name manually here.
 488          */
 489         dev_set_name(&rdi->ibdev.dev, fmt, name, unit);
 490         strlcpy(rdi->ibdev.name, dev_name(&rdi->ibdev.dev), IB_DEVICE_NAME_MAX);
 491 }
 492
 493 /**
 494  * rvt_get_ibdev_name - return the IB name
 495  * @rdi: rdmavt device
 496  *
 497  * Return the registered name of the device.
 498  */
 499 static inline const char *rvt_get_ibdev_name(const struct rvt_dev_info *rdi)
 500 {
 501         return dev_name(&rdi->ibdev.dev);
 502 }
 503
 504 static inline struct rvt_pd *ibpd_to_rvtpd(struct ib_pd *ibpd)
 505 {
 506         return container_of(ibpd, struct rvt_pd, ibpd);
 507 }
 508
 509 static inline struct rvt_ah *ibah_to_rvtah(struct ib_ah *ibah)
 510 {
 511         return container_of(ibah, struct rvt_ah, ibah);
 512 }
 513
 514 static inline struct rvt_dev_info *ib_to_rvt(struct ib_device *ibdev)
 515 {
 516         return  container_of(ibdev, struct rvt_dev_info, ibdev);
 517 }
 518
 519 static inline unsigned rvt_get_npkeys(struct rvt_dev_info *rdi)
 520 {
 521         /*
 522          * All ports have same number of pkeys.
 523          */
 524         return rdi->dparms.npkeys;
 525 }
 526
 527 /*
 528  * Return the max atomic suitable for determining
 529  * the size of the ack ring buffer in a QP.
 530  */
 531 static inline unsigned int rvt_max_atomic(struct rvt_dev_info *rdi)
 532 {
 533         return rdi->dparms.max_rdma_atomic +
 534                 rdi->dparms.extra_rdma_atomic + 1;
 535 }
 536
 537 static inline unsigned int rvt_size_atomic(struct rvt_dev_info *rdi)
 538 {
 539         return rdi->dparms.max_rdma_atomic +
 540                 rdi->dparms.extra_rdma_atomic;
 541 }
 542
 543 /*
 544  * Return the indexed PKEY from the port PKEY table.
 545  */
 546 static inline u16 rvt_get_pkey(struct rvt_dev_info *rdi,
 547                                int port_index,
 548                                unsigned index)
 549 {
 550         if (index >= rvt_get_npkeys(rdi))
 551                 return 0;
 552         else
 553                 return rdi->ports[port_index]->pkey_table[index];
 554 }
 555
 556 struct rvt_dev_info *rvt_alloc_device(size_t size, int nports);
 557 void rvt_dealloc_device(struct rvt_dev_info *rdi);
 558 int rvt_register_device(struct rvt_dev_info *rvd);
 559 void rvt_unregister_device(struct rvt_dev_info *rvd);
 560 int rvt_check_ah(struct ib_device *ibdev, struct rdma_ah_attr *ah_attr);
 561 int rvt_init_port(struct rvt_dev_info *rdi, struct rvt_ibport *port,
 562                   int port_index, u16 *pkey_table);
 563 int rvt_fast_reg_mr(struct rvt_qp *qp, struct ib_mr *ibmr, u32 key,
 564                     int access);
 565 int rvt_invalidate_rkey(struct rvt_qp *qp, u32 rkey);
 566 int rvt_rkey_ok(struct rvt_qp *qp, struct rvt_sge *sge,
 567                 u32 len, u64 vaddr, u32 rkey, int acc);
 568 int rvt_lkey_ok(struct rvt_lkey_table *rkt, struct rvt_pd *pd,
 569                 struct rvt_sge *isge, struct rvt_sge *last_sge,
 570                 struct ib_sge *sge, int acc);
 571 struct rvt_mcast *rvt_mcast_find(struct rvt_ibport *ibp, union ib_gid *mgid,
 572                                  u16 lid);
 573
 574 #endif          /* DEF_RDMA_VT_H */