1 #ifndef DEF_RDMA_VT_H 2 #define DEF_RDMA_VT_H 3 4 /* 5 * Copyright(c) 2015 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 /* 64 * For some of the IBTA objects there will likely be some 65 * initializations required. We need flags to determine whether it is OK 66 * for rdmavt to do this or not. This does not imply any functions of a 67 * partiuclar IBTA object are overridden. 68 */ 69 #define RVT_FLAG_MR_INIT_DRIVER BIT(1) 70 #define RVT_FLAG_QP_INIT_DRIVER BIT(2) 71 #define RVT_FLAG_CQ_INIT_DRIVER BIT(3) 72 73 #define RVT_MAX_PKEY_VALUES 16 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 u32 pma_sample_start; 89 u32 pma_sample_interval; 90 __be16 pma_counter_select[5]; 91 u16 pma_tag; 92 u16 mkey_lease_period; 93 u16 sm_lid; 94 u8 sm_sl; 95 u8 mkeyprot; 96 u8 subnet_timeout; 97 u8 vl_high_limit; 98 99 /* 100 * Driver is expected to keep these up to date. These 101 * counters are informational only and not required to be 102 * completely accurate. 103 */ 104 u64 n_rc_resends; 105 u64 n_seq_naks; 106 u64 n_rdma_seq; 107 u64 n_rnr_naks; 108 u64 n_other_naks; 109 u64 n_loop_pkts; 110 u64 n_pkt_drops; 111 u64 n_vl15_dropped; 112 u64 n_rc_timeouts; 113 u64 n_dmawait; 114 u64 n_unaligned; 115 u64 n_rc_dupreq; 116 u64 n_rc_seqnak; 117 u16 pkey_violations; 118 u16 qkey_violations; 119 u16 mkey_violations; 120 121 /* Hot-path per CPU counters to avoid cacheline trading to update */ 122 u64 z_rc_acks; 123 u64 z_rc_qacks; 124 u64 z_rc_delayed_comp; 125 u64 __percpu *rc_acks; 126 u64 __percpu *rc_qacks; 127 u64 __percpu *rc_delayed_comp; 128 129 void *priv; /* driver private data */ 130 131 /* 132 * The pkey table is allocated and maintained by the driver. Drivers 133 * need to have access to this before registering with rdmav. However 134 * rdmavt will need access to it so drivers need to proviee this during 135 * the attach port API call. 136 */ 137 u16 *pkey_table; 138 139 /* TODO: Move sm_ah and smi_ah into here as well*/ 140 }; 141 142 #define RVT_CQN_MAX 16 /* maximum length of cq name */ 143 144 /* 145 * Things that are driver specific, module parameters in hfi1 and qib 146 */ 147 struct rvt_driver_params { 148 /* 149 * driver required fields: 150 * node_guid 151 * phys_port_cnt 152 * dma_device 153 * owner 154 * driver optional fields (rvt will provide generic value if blank): 155 * name 156 * node_desc 157 * rvt fields, driver value ignored: 158 * uverbs_abi_ver 159 * node_type 160 * num_comp_vectors 161 * uverbs_cmd_mask 162 */ 163 struct ib_device_attr props; 164 165 /* 166 * Drivers will need to support a number of notifications to rvt in 167 * accordance with certain events. This structure should contain a mask 168 * of the supported events. Such events that the rvt may need to know 169 * about include: 170 * port errors 171 * port active 172 * lid change 173 * sm change 174 * client reregister 175 * pkey change 176 * 177 * There may also be other events that the rvt layers needs to know 178 * about this is not an exhaustive list. Some events though rvt does not 179 * need to rely on the driver for such as completion queue error. 180 */ 181 int rvt_signal_supported; 182 183 /* 184 * Anything driver specific that is not covered by props 185 * For instance special module parameters. Goes here. 186 */ 187 unsigned int lkey_table_size; 188 unsigned int qp_table_size; 189 int qpn_start; 190 int qpn_inc; 191 int qpn_res_start; 192 int qpn_res_end; 193 int nports; 194 int npkeys; 195 u8 qos_shift; 196 char cq_name[RVT_CQN_MAX]; 197 int node; 198 int max_rdma_atomic; 199 int psn_mask; 200 int psn_shift; 201 int psn_modify_mask; 202 }; 203 204 /* Protection domain */ 205 struct rvt_pd { 206 struct ib_pd ibpd; 207 int user; /* non-zero if created from user space */ 208 }; 209 210 /* Address handle */ 211 struct rvt_ah { 212 struct ib_ah ibah; 213 struct ib_ah_attr attr; 214 atomic_t refcount; 215 u8 vl; 216 u8 log_pmtu; 217 }; 218 219 struct rvt_dev_info; 220 struct rvt_driver_provided { 221 /* 222 * The work to create port files in /sys/class Infiniband is different 223 * depending on the driver. This should not be extracted away and 224 * instead drivers are responsible for setting the correct callback for 225 * this. 226 */ 227 228 /* -------------------*/ 229 /* Required functions */ 230 /* -------------------*/ 231 int (*port_callback)(struct ib_device *, u8, struct kobject *); 232 const char * (*get_card_name)(struct rvt_dev_info *rdi); 233 struct pci_dev * (*get_pci_dev)(struct rvt_dev_info *rdi); 234 unsigned (*free_all_qps)(struct rvt_dev_info *rdi); 235 void * (*qp_priv_alloc)(struct rvt_dev_info *rdi, struct rvt_qp *qp, 236 gfp_t gfp); 237 void (*qp_priv_free)(struct rvt_dev_info *rdi, struct rvt_qp *qp); 238 void (*notify_qp_reset)(struct rvt_qp *qp); 239 void (*schedule_send)(struct rvt_qp *qp); 240 void (*do_send)(struct rvt_qp *qp); 241 int (*get_pmtu_from_attr)(struct rvt_dev_info *rdi, struct rvt_qp *qp, 242 struct ib_qp_attr *attr); 243 void (*flush_qp_waiters)(struct rvt_qp *qp); 244 void (*stop_send_queue)(struct rvt_qp *qp); 245 void (*quiesce_qp)(struct rvt_qp *qp); 246 void (*notify_error_qp)(struct rvt_qp *qp); 247 u32 (*mtu_from_qp)(struct rvt_dev_info *rdi, struct rvt_qp *qp, 248 u32 pmtu); 249 int (*mtu_to_path_mtu)(u32 mtu); 250 251 /*--------------------*/ 252 /* Optional functions */ 253 /*--------------------*/ 254 int (*check_ah)(struct ib_device *, struct ib_ah_attr *); 255 void (*notify_new_ah)(struct ib_device *, struct ib_ah_attr *, 256 struct rvt_ah *); 257 int (*alloc_qpn)(struct rvt_dev_info *rdi, struct rvt_qpn_table *qpt, 258 enum ib_qp_type type, u8 port, gfp_t gfp); 259 /** 260 * Return 0 if modification is valid, -errno otherwise 261 */ 262 int (*check_modify_qp)(struct rvt_qp *qp, struct ib_qp_attr *attr, 263 int attr_mask, struct ib_udata *udata); 264 void (*modify_qp)(struct rvt_qp *qp, struct ib_qp_attr *attr, 265 int attr_mask, struct ib_udata *udata); 266 }; 267 268 struct rvt_dev_info { 269 struct ib_device ibdev; /* Keep this first. Nothing above here */ 270 271 /* 272 * Prior to calling for registration the driver will be responsible for 273 * allocating space for this structure. 274 * 275 * The driver will also be responsible for filling in certain members of 276 * dparms.props. The driver needs to fill in dparms exactly as it would 277 * want values reported to a ULP. This will be returned to the caller 278 * in rdmavt's device. The driver should also therefore refrain from 279 * modifying this directly after registration with rdmavt. 280 */ 281 282 /* Driver specific properties */ 283 struct rvt_driver_params dparms; 284 285 struct rvt_mregion __rcu *dma_mr; 286 struct rvt_lkey_table lkey_table; 287 288 /* Driver specific helper functions */ 289 struct rvt_driver_provided driver_f; 290 291 /* Internal use */ 292 int n_pds_allocated; 293 spinlock_t n_pds_lock; /* Protect pd allocated count */ 294 295 int n_ahs_allocated; 296 spinlock_t n_ahs_lock; /* Protect ah allocated count */ 297 298 int flags; 299 struct rvt_ibport **ports; 300 301 /* QP */ 302 struct rvt_qp_ibdev *qp_dev; 303 u32 n_qps_allocated; /* number of QPs allocated for device */ 304 spinlock_t n_qps_lock; /* keep track of number of qps */ 305 306 /* memory maps */ 307 struct list_head pending_mmaps; 308 spinlock_t mmap_offset_lock; /* protect mmap_offset */ 309 u32 mmap_offset; 310 spinlock_t pending_lock; /* protect pending mmap list */ 311 312 /* CQ */ 313 struct kthread_worker *worker; /* per device cq worker */ 314 u32 n_cqs_allocated; /* number of CQs allocated for device */ 315 spinlock_t n_cqs_lock; /* protect count of in use cqs */ 316 317 /* Multicast */ 318 u32 n_mcast_grps_allocated; /* number of mcast groups allocated */ 319 spinlock_t n_mcast_grps_lock; 320 321 }; 322 323 static inline struct rvt_pd *ibpd_to_rvtpd(struct ib_pd *ibpd) 324 { 325 return container_of(ibpd, struct rvt_pd, ibpd); 326 } 327 328 static inline struct rvt_ah *ibah_to_rvtah(struct ib_ah *ibah) 329 { 330 return container_of(ibah, struct rvt_ah, ibah); 331 } 332 333 static inline struct rvt_dev_info *ib_to_rvt(struct ib_device *ibdev) 334 { 335 return container_of(ibdev, struct rvt_dev_info, ibdev); 336 } 337 338 static inline struct rvt_srq *ibsrq_to_rvtsrq(struct ib_srq *ibsrq) 339 { 340 return container_of(ibsrq, struct rvt_srq, ibsrq); 341 } 342 343 static inline struct rvt_qp *ibqp_to_rvtqp(struct ib_qp *ibqp) 344 { 345 return container_of(ibqp, struct rvt_qp, ibqp); 346 } 347 348 static inline unsigned rvt_get_npkeys(struct rvt_dev_info *rdi) 349 { 350 /* 351 * All ports have same number of pkeys. 352 */ 353 return rdi->dparms.npkeys; 354 } 355 356 /* 357 * Return the indexed PKEY from the port PKEY table. 358 */ 359 static inline u16 rvt_get_pkey(struct rvt_dev_info *rdi, 360 int port_index, 361 unsigned index) 362 { 363 if (index >= rvt_get_npkeys(rdi)) 364 return 0; 365 else 366 return rdi->ports[port_index]->pkey_table[index]; 367 } 368 369 /** 370 * rvt_lookup_qpn - return the QP with the given QPN 371 * @ibp: the ibport 372 * @qpn: the QP number to look up 373 * 374 * The caller must hold the rcu_read_lock(), and keep the lock until 375 * the returned qp is no longer in use. 376 */ 377 /* TODO: Remove this and put in rdmavt/qp.h when no longer needed by drivers */ 378 static inline struct rvt_qp *rvt_lookup_qpn(struct rvt_dev_info *rdi, 379 struct rvt_ibport *rvp, 380 u32 qpn) __must_hold(RCU) 381 { 382 struct rvt_qp *qp = NULL; 383 384 if (unlikely(qpn <= 1)) { 385 qp = rcu_dereference(rvp->qp[qpn]); 386 } else { 387 u32 n = hash_32(qpn, rdi->qp_dev->qp_table_bits); 388 389 for (qp = rcu_dereference(rdi->qp_dev->qp_table[n]); qp; 390 qp = rcu_dereference(qp->next)) 391 if (qp->ibqp.qp_num == qpn) 392 break; 393 } 394 return qp; 395 } 396 397 struct rvt_dev_info *rvt_alloc_device(size_t size, int nports); 398 int rvt_register_device(struct rvt_dev_info *rvd); 399 void rvt_unregister_device(struct rvt_dev_info *rvd); 400 int rvt_check_ah(struct ib_device *ibdev, struct ib_ah_attr *ah_attr); 401 int rvt_init_port(struct rvt_dev_info *rdi, struct rvt_ibport *port, 402 int portnum, u16 *pkey_table); 403 int rvt_rkey_ok(struct rvt_qp *qp, struct rvt_sge *sge, 404 u32 len, u64 vaddr, u32 rkey, int acc); 405 int rvt_lkey_ok(struct rvt_lkey_table *rkt, struct rvt_pd *pd, 406 struct rvt_sge *isge, struct ib_sge *sge, int acc); 407 int rvt_mmap(struct ib_ucontext *context, struct vm_area_struct *vma); 408 void rvt_release_mmap_info(struct kref *ref); 409 struct rvt_mmap_info *rvt_create_mmap_info(struct rvt_dev_info *rdi, 410 u32 size, 411 struct ib_ucontext *context, 412 void *obj); 413 void rvt_update_mmap_info(struct rvt_dev_info *rdi, struct rvt_mmap_info *ip, 414 u32 size, void *obj); 415 int rvt_reg_mr(struct rvt_qp *qp, struct ib_reg_wr *wr); 416 struct rvt_mcast *rvt_mcast_find(struct rvt_ibport *ibp, union ib_gid *mgid); 417 418 /* Temporary export */ 419 void rvt_reset_qp(struct rvt_dev_info *rdi, struct rvt_qp *qp, 420 enum ib_qp_type type); 421 422 #endif /* DEF_RDMA_VT_H */ 423