5 * Copyright(c) 2016 - 2018 Intel Corporation.
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.
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.
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.
23 * Redistribution and use in source and binary forms, with or without
24 * modification, are permitted provided that the following conditions
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
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.
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.
52 * Structure that low level drivers will populate in order to register with the
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>
64 #define RVT_MAX_PKEY_VALUES 16
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 */
72 struct list_head list;
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 */
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 */
90 u32 pma_sample_interval;
91 __be16 pma_counter_select[5];
93 u16 mkey_lease_period;
101 * Driver is expected to keep these up to date. These
102 * counters are informational only and not required to be
103 * completely accurate.
122 /* Hot-path per CPU counters to avoid cacheline trading to update */
125 u64 z_rc_delayed_comp;
126 u64 __percpu *rc_acks;
127 u64 __percpu *rc_qacks;
128 u64 __percpu *rc_delayed_comp;
130 void *priv; /* driver private data */
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.
140 struct rvt_ah *sm_ah;
143 * Keep a list of traps that have not been repressed. They will be
144 * resent based on trap_timer.
146 struct trap_list trap_lists[RVT_MAX_TRAP_LISTS];
147 struct timer_list trap_timer;
150 #define RVT_CQN_MAX 16 /* maximum length of cq name */
153 * Things that are driver specific, module parameters in hfi1 and qib
155 struct rvt_driver_params {
156 struct ib_device_attr props;
159 * Anything driver specific that is not covered by props
160 * For instance special module parameters. Goes here.
162 unsigned int lkey_table_size;
163 unsigned int qp_table_size;
178 u8 reserved_operations;
181 /* Protection domain */
190 struct rdma_ah_attr attr;
198 struct rvt_driver_provided {
200 * Which functions are required depends on which verbs rdmavt is
201 * providing and which verbs the driver is overriding. See
202 * check_support() for details.
205 /* hot path calldowns in a single cacheline */
208 * Give the driver a notice that there is send work to do. It is up to
209 * the driver to generally push the packets out, this just queues the
210 * work with the driver. There are two variants here. The no_lock
211 * version requires the s_lock not to be held. The other assumes the
214 void (*schedule_send)(struct rvt_qp *qp);
215 void (*schedule_send_no_lock)(struct rvt_qp *qp);
218 * Driver specific work request setup and checking.
219 * This function is allowed to perform any setup, checks, or
220 * adjustments required to the SWQE in order to be usable by
221 * underlying protocols. This includes private data structure
224 int (*setup_wqe)(struct rvt_qp *qp, struct rvt_swqe *wqe,
228 * Sometimes rdmavt needs to kick the driver's send progress. That is
229 * done by this call back.
231 void (*do_send)(struct rvt_qp *qp);
233 /* Passed to ib core registration. Callback to create syfs files */
234 int (*port_callback)(struct ib_device *, u8, struct kobject *);
237 * Returns a pointer to the undelying hardware's PCI device. This is
238 * used to display information as to what hardware is being referenced
239 * in an output message
241 struct pci_dev * (*get_pci_dev)(struct rvt_dev_info *rdi);
244 * Allocate a private queue pair data structure for driver specific
245 * information which is opaque to rdmavt. Errors are returned via
246 * ERR_PTR(err). The driver is free to return NULL or a valid
249 void * (*qp_priv_alloc)(struct rvt_dev_info *rdi, struct rvt_qp *qp);
252 * Free the driver's private qp structure.
254 void (*qp_priv_free)(struct rvt_dev_info *rdi, struct rvt_qp *qp);
257 * Inform the driver the particular qp in quesiton has been reset so
258 * that it can clean up anything it needs to.
260 void (*notify_qp_reset)(struct rvt_qp *qp);
263 * Get a path mtu from the driver based on qp attributes.
265 int (*get_pmtu_from_attr)(struct rvt_dev_info *rdi, struct rvt_qp *qp,
266 struct ib_qp_attr *attr);
269 * Notify driver that it needs to flush any outstanding IO requests that
270 * are waiting on a qp.
272 void (*flush_qp_waiters)(struct rvt_qp *qp);
275 * Notify driver to stop its queue of sending packets. Nothing else
276 * should be posted to the queue pair after this has been called.
278 void (*stop_send_queue)(struct rvt_qp *qp);
281 * Have the drivr drain any in progress operations
283 void (*quiesce_qp)(struct rvt_qp *qp);
286 * Inform the driver a qp has went to error state.
288 void (*notify_error_qp)(struct rvt_qp *qp);
291 * Get an MTU for a qp.
293 u32 (*mtu_from_qp)(struct rvt_dev_info *rdi, struct rvt_qp *qp,
296 * Convert an mtu to a path mtu
298 int (*mtu_to_path_mtu)(u32 mtu);
301 * Get the guid of a port in big endian byte order
303 int (*get_guid_be)(struct rvt_dev_info *rdi, struct rvt_ibport *rvp,
304 int guid_index, __be64 *guid);
307 * Query driver for the state of the port.
309 int (*query_port_state)(struct rvt_dev_info *rdi, u8 port_num,
310 struct ib_port_attr *props);
313 * Tell driver to shutdown a port
315 int (*shut_down_port)(struct rvt_dev_info *rdi, u8 port_num);
317 /* Tell driver to send a trap for changed port capabilities */
318 void (*cap_mask_chg)(struct rvt_dev_info *rdi, u8 port_num);
321 * The following functions can be safely ignored completely. Any use of
322 * these is checked for NULL before blindly calling. Rdmavt should also
323 * be functional if drivers omit these.
326 /* Called to inform the driver that all qps should now be freed. */
327 unsigned (*free_all_qps)(struct rvt_dev_info *rdi);
329 /* Driver specific AH validation */
330 int (*check_ah)(struct ib_device *, struct rdma_ah_attr *);
332 /* Inform the driver a new AH has been created */
333 void (*notify_new_ah)(struct ib_device *, struct rdma_ah_attr *,
336 /* Let the driver pick the next queue pair number*/
337 int (*alloc_qpn)(struct rvt_dev_info *rdi, struct rvt_qpn_table *qpt,
338 enum ib_qp_type type, u8 port_num);
340 /* Determine if its safe or allowed to modify the qp */
341 int (*check_modify_qp)(struct rvt_qp *qp, struct ib_qp_attr *attr,
342 int attr_mask, struct ib_udata *udata);
344 /* Driver specific QP modification/notification-of */
345 void (*modify_qp)(struct rvt_qp *qp, struct ib_qp_attr *attr,
346 int attr_mask, struct ib_udata *udata);
348 /* Notify driver a mad agent has been created */
349 void (*notify_create_mad_agent)(struct rvt_dev_info *rdi, int port_idx);
351 /* Notify driver a mad agent has been removed */
352 void (*notify_free_mad_agent)(struct rvt_dev_info *rdi, int port_idx);
354 /* Notify driver to restart rc */
355 void (*notify_restart_rc)(struct rvt_qp *qp, u32 psn, int wait);
357 /* Get and return CPU to pin CQ processing thread */
358 int (*comp_vect_cpu_lookup)(struct rvt_dev_info *rdi, int comp_vect);
361 struct rvt_dev_info {
362 struct ib_device ibdev; /* Keep this first. Nothing above here */
365 * Prior to calling for registration the driver will be responsible for
366 * allocating space for this structure.
368 * The driver will also be responsible for filling in certain members of
369 * dparms.props. The driver needs to fill in dparms exactly as it would
370 * want values reported to a ULP. This will be returned to the caller
371 * in rdmavt's device. The driver should also therefore refrain from
372 * modifying this directly after registration with rdmavt.
375 /* Driver specific properties */
376 struct rvt_driver_params dparms;
378 /* post send table */
379 const struct rvt_operation_params *post_parms;
381 /* Driver specific helper functions */
382 struct rvt_driver_provided driver_f;
384 struct rvt_mregion __rcu *dma_mr;
385 struct rvt_lkey_table lkey_table;
389 spinlock_t n_pds_lock; /* Protect pd allocated count */
392 spinlock_t n_ahs_lock; /* Protect ah allocated count */
394 u32 n_srqs_allocated;
395 spinlock_t n_srqs_lock; /* Protect srqs allocated count */
398 struct rvt_ibport **ports;
401 struct rvt_qp_ibdev *qp_dev;
402 u32 n_qps_allocated; /* number of QPs allocated for device */
403 u32 n_rc_qps; /* number of RC QPs allocated for device */
404 u32 busy_jiffies; /* timeout scaling based on RC QP count */
405 spinlock_t n_qps_lock; /* protect qps, rc qps and busy jiffy counts */
408 struct list_head pending_mmaps;
409 spinlock_t mmap_offset_lock; /* protect mmap_offset */
411 spinlock_t pending_lock; /* protect pending mmap list */
414 u32 n_cqs_allocated; /* number of CQs allocated for device */
415 spinlock_t n_cqs_lock; /* protect count of in use cqs */
418 u32 n_mcast_grps_allocated; /* number of mcast groups allocated */
419 spinlock_t n_mcast_grps_lock;
424 * rvt_set_ibdev_name - Craft an IB device name from client info
425 * @rdi: pointer to the client rvt_dev_info structure
426 * @name: client specific name
427 * @unit: client specific unit number.
429 static inline void rvt_set_ibdev_name(struct rvt_dev_info *rdi,
430 const char *fmt, const char *name,
434 * FIXME: rvt and its users want to touch the ibdev before
435 * registration and have things like the name work. We don't have the
436 * infrastructure in the core to support this directly today, hack it
437 * to work by setting the name manually here.
439 dev_set_name(&rdi->ibdev.dev, fmt, name, unit);
440 strlcpy(rdi->ibdev.name, dev_name(&rdi->ibdev.dev), IB_DEVICE_NAME_MAX);
444 * rvt_get_ibdev_name - return the IB name
445 * @rdi: rdmavt device
447 * Return the registered name of the device.
449 static inline const char *rvt_get_ibdev_name(const struct rvt_dev_info *rdi)
451 return dev_name(&rdi->ibdev.dev);
454 static inline struct rvt_pd *ibpd_to_rvtpd(struct ib_pd *ibpd)
456 return container_of(ibpd, struct rvt_pd, ibpd);
459 static inline struct rvt_ah *ibah_to_rvtah(struct ib_ah *ibah)
461 return container_of(ibah, struct rvt_ah, ibah);
464 static inline struct rvt_dev_info *ib_to_rvt(struct ib_device *ibdev)
466 return container_of(ibdev, struct rvt_dev_info, ibdev);
469 static inline struct rvt_srq *ibsrq_to_rvtsrq(struct ib_srq *ibsrq)
471 return container_of(ibsrq, struct rvt_srq, ibsrq);
474 static inline struct rvt_qp *ibqp_to_rvtqp(struct ib_qp *ibqp)
476 return container_of(ibqp, struct rvt_qp, ibqp);
479 static inline unsigned rvt_get_npkeys(struct rvt_dev_info *rdi)
482 * All ports have same number of pkeys.
484 return rdi->dparms.npkeys;
488 * Return the max atomic suitable for determining
489 * the size of the ack ring buffer in a QP.
491 static inline unsigned int rvt_max_atomic(struct rvt_dev_info *rdi)
493 return rdi->dparms.max_rdma_atomic + 1;
497 * Return the indexed PKEY from the port PKEY table.
499 static inline u16 rvt_get_pkey(struct rvt_dev_info *rdi,
503 if (index >= rvt_get_npkeys(rdi))
506 return rdi->ports[port_index]->pkey_table[index];
510 * rvt_lookup_qpn - return the QP with the given QPN
512 * @qpn: the QP number to look up
514 * The caller must hold the rcu_read_lock(), and keep the lock until
515 * the returned qp is no longer in use.
517 /* TODO: Remove this and put in rdmavt/qp.h when no longer needed by drivers */
518 static inline struct rvt_qp *rvt_lookup_qpn(struct rvt_dev_info *rdi,
519 struct rvt_ibport *rvp,
520 u32 qpn) __must_hold(RCU)
522 struct rvt_qp *qp = NULL;
524 if (unlikely(qpn <= 1)) {
525 qp = rcu_dereference(rvp->qp[qpn]);
527 u32 n = hash_32(qpn, rdi->qp_dev->qp_table_bits);
529 for (qp = rcu_dereference(rdi->qp_dev->qp_table[n]); qp;
530 qp = rcu_dereference(qp->next))
531 if (qp->ibqp.qp_num == qpn)
538 * rvt_mod_retry_timer - mod a retry timer
540 * Modify a potentially already running retry timer
542 static inline void rvt_mod_retry_timer(struct rvt_qp *qp)
544 struct ib_qp *ibqp = &qp->ibqp;
545 struct rvt_dev_info *rdi = ib_to_rvt(ibqp->device);
547 lockdep_assert_held(&qp->s_lock);
548 qp->s_flags |= RVT_S_TIMER;
549 /* 4.096 usec. * (1 << qp->timeout) */
550 mod_timer(&qp->s_timer, jiffies + qp->timeout_jiffies +
554 struct rvt_dev_info *rvt_alloc_device(size_t size, int nports);
555 void rvt_dealloc_device(struct rvt_dev_info *rdi);
556 int rvt_register_device(struct rvt_dev_info *rvd, u32 driver_id);
557 void rvt_unregister_device(struct rvt_dev_info *rvd);
558 int rvt_check_ah(struct ib_device *ibdev, struct rdma_ah_attr *ah_attr);
559 int rvt_init_port(struct rvt_dev_info *rdi, struct rvt_ibport *port,
560 int port_index, u16 *pkey_table);
561 int rvt_fast_reg_mr(struct rvt_qp *qp, struct ib_mr *ibmr, u32 key,
563 int rvt_invalidate_rkey(struct rvt_qp *qp, u32 rkey);
564 int rvt_rkey_ok(struct rvt_qp *qp, struct rvt_sge *sge,
565 u32 len, u64 vaddr, u32 rkey, int acc);
566 int rvt_lkey_ok(struct rvt_lkey_table *rkt, struct rvt_pd *pd,
567 struct rvt_sge *isge, struct rvt_sge *last_sge,
568 struct ib_sge *sge, int acc);
569 struct rvt_mcast *rvt_mcast_find(struct rvt_ibport *ibp, union ib_gid *mgid,
572 #endif /* DEF_RDMA_VT_H */