2 * Copyright(c) 2015, 2016 Intel Corporation.
4 * This file is provided under a dual BSD/GPLv2 license. When using or
5 * redistributing this file, you may do so under either license.
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of version 2 of the GNU General Public License as
11 * published by the Free Software Foundation.
13 * This program is distributed in the hope that it will be useful, but
14 * WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * General Public License for more details.
20 * Redistribution and use in source and binary forms, with or without
21 * modification, are permitted provided that the following conditions
24 * - Redistributions of source code must retain the above copyright
25 * notice, this list of conditions and the following disclaimer.
26 * - Redistributions in binary form must reproduce the above copyright
27 * notice, this list of conditions and the following disclaimer in
28 * the documentation and/or other materials provided with the
30 * - Neither the name of Intel Corporation nor the names of its
31 * contributors may be used to endorse or promote products derived
32 * from this software without specific prior written permission.
34 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
35 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
36 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
37 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
38 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
39 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
40 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
41 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
42 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
43 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
44 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
48 #include <linux/err.h>
49 #include <linux/vmalloc.h>
50 #include <linux/hash.h>
51 #include <linux/module.h>
52 #include <linux/seq_file.h>
53 #include <rdma/rdma_vt.h>
54 #include <rdma/rdmavt_qp.h>
55 #include <rdma/ib_verbs.h>
60 #include "verbs_txreq.h"
62 unsigned int hfi1_qp_table_size
= 256;
63 module_param_named(qp_table_size
, hfi1_qp_table_size
, uint
, S_IRUGO
);
64 MODULE_PARM_DESC(qp_table_size
, "QP table size");
66 static void flush_tx_list(struct rvt_qp
*qp
);
67 static int iowait_sleep(
68 struct sdma_engine
*sde
,
70 struct sdma_txreq
*stx
,
72 static void iowait_wakeup(struct iowait
*wait
, int reason
);
73 static void iowait_sdma_drained(struct iowait
*wait
);
74 static void qp_pio_drain(struct rvt_qp
*qp
);
76 static inline unsigned mk_qpn(struct rvt_qpn_table
*qpt
,
77 struct rvt_qpn_map
*map
, unsigned off
)
79 return (map
- qpt
->map
) * RVT_BITS_PER_PAGE
+ off
;
82 const struct rvt_operation_params hfi1_post_parms
[RVT_OPERATION_MAX
] = {
83 [IB_WR_RDMA_WRITE
] = {
84 .length
= sizeof(struct ib_rdma_wr
),
85 .qpt_support
= BIT(IB_QPT_UC
) | BIT(IB_QPT_RC
),
89 .length
= sizeof(struct ib_rdma_wr
),
90 .qpt_support
= BIT(IB_QPT_RC
),
91 .flags
= RVT_OPERATION_ATOMIC
,
94 [IB_WR_ATOMIC_CMP_AND_SWP
] = {
95 .length
= sizeof(struct ib_atomic_wr
),
96 .qpt_support
= BIT(IB_QPT_RC
),
97 .flags
= RVT_OPERATION_ATOMIC
| RVT_OPERATION_ATOMIC_SGE
,
100 [IB_WR_ATOMIC_FETCH_AND_ADD
] = {
101 .length
= sizeof(struct ib_atomic_wr
),
102 .qpt_support
= BIT(IB_QPT_RC
),
103 .flags
= RVT_OPERATION_ATOMIC
| RVT_OPERATION_ATOMIC_SGE
,
106 [IB_WR_RDMA_WRITE_WITH_IMM
] = {
107 .length
= sizeof(struct ib_rdma_wr
),
108 .qpt_support
= BIT(IB_QPT_UC
) | BIT(IB_QPT_RC
),
112 .length
= sizeof(struct ib_send_wr
),
113 .qpt_support
= BIT(IB_QPT_UD
) | BIT(IB_QPT_SMI
) | BIT(IB_QPT_GSI
) |
114 BIT(IB_QPT_UC
) | BIT(IB_QPT_RC
),
117 [IB_WR_SEND_WITH_IMM
] = {
118 .length
= sizeof(struct ib_send_wr
),
119 .qpt_support
= BIT(IB_QPT_UD
) | BIT(IB_QPT_SMI
) | BIT(IB_QPT_GSI
) |
120 BIT(IB_QPT_UC
) | BIT(IB_QPT_RC
),
124 .length
= sizeof(struct ib_reg_wr
),
125 .qpt_support
= BIT(IB_QPT_UC
) | BIT(IB_QPT_RC
),
126 .flags
= RVT_OPERATION_LOCAL
,
129 [IB_WR_LOCAL_INV
] = {
130 .length
= sizeof(struct ib_send_wr
),
131 .qpt_support
= BIT(IB_QPT_UC
) | BIT(IB_QPT_RC
),
132 .flags
= RVT_OPERATION_LOCAL
,
135 [IB_WR_SEND_WITH_INV
] = {
136 .length
= sizeof(struct ib_send_wr
),
137 .qpt_support
= BIT(IB_QPT_RC
),
142 static void flush_tx_list(struct rvt_qp
*qp
)
144 struct hfi1_qp_priv
*priv
= qp
->priv
;
146 while (!list_empty(&priv
->s_iowait
.tx_head
)) {
147 struct sdma_txreq
*tx
;
149 tx
= list_first_entry(
150 &priv
->s_iowait
.tx_head
,
153 list_del_init(&tx
->list
);
155 container_of(tx
, struct verbs_txreq
, txreq
));
159 static void flush_iowait(struct rvt_qp
*qp
)
161 struct hfi1_qp_priv
*priv
= qp
->priv
;
163 seqlock_t
*lock
= priv
->s_iowait
.lock
;
167 write_seqlock_irqsave(lock
, flags
);
168 if (!list_empty(&priv
->s_iowait
.list
)) {
169 list_del_init(&priv
->s_iowait
.list
);
170 priv
->s_iowait
.lock
= NULL
;
173 write_sequnlock_irqrestore(lock
, flags
);
176 static inline int opa_mtu_enum_to_int(int mtu
)
179 case OPA_MTU_8192
: return 8192;
180 case OPA_MTU_10240
: return 10240;
186 * This function is what we would push to the core layer if we wanted to be a
187 * "first class citizen". Instead we hide this here and rely on Verbs ULPs
188 * to blindly pass the MTU enum value from the PathRecord to us.
190 static inline int verbs_mtu_enum_to_int(struct ib_device
*dev
, enum ib_mtu mtu
)
194 /* Constraining 10KB packets to 8KB packets */
195 if (mtu
== (enum ib_mtu
)OPA_MTU_10240
)
197 val
= opa_mtu_enum_to_int((int)mtu
);
200 return ib_mtu_enum_to_int(mtu
);
203 int hfi1_check_modify_qp(struct rvt_qp
*qp
, struct ib_qp_attr
*attr
,
204 int attr_mask
, struct ib_udata
*udata
)
206 struct ib_qp
*ibqp
= &qp
->ibqp
;
207 struct hfi1_ibdev
*dev
= to_idev(ibqp
->device
);
208 struct hfi1_devdata
*dd
= dd_from_dev(dev
);
211 if (attr_mask
& IB_QP_AV
) {
212 sc
= ah_to_sc(ibqp
->device
, &attr
->ah_attr
);
216 if (!qp_to_sdma_engine(qp
, sc
) &&
217 dd
->flags
& HFI1_HAS_SEND_DMA
)
220 if (!qp_to_send_context(qp
, sc
))
224 if (attr_mask
& IB_QP_ALT_PATH
) {
225 sc
= ah_to_sc(ibqp
->device
, &attr
->alt_ah_attr
);
229 if (!qp_to_sdma_engine(qp
, sc
) &&
230 dd
->flags
& HFI1_HAS_SEND_DMA
)
233 if (!qp_to_send_context(qp
, sc
))
240 void hfi1_modify_qp(struct rvt_qp
*qp
, struct ib_qp_attr
*attr
,
241 int attr_mask
, struct ib_udata
*udata
)
243 struct ib_qp
*ibqp
= &qp
->ibqp
;
244 struct hfi1_qp_priv
*priv
= qp
->priv
;
246 if (attr_mask
& IB_QP_AV
) {
247 priv
->s_sc
= ah_to_sc(ibqp
->device
, &qp
->remote_ah_attr
);
248 priv
->s_sde
= qp_to_sdma_engine(qp
, priv
->s_sc
);
249 priv
->s_sendcontext
= qp_to_send_context(qp
, priv
->s_sc
);
252 if (attr_mask
& IB_QP_PATH_MIG_STATE
&&
253 attr
->path_mig_state
== IB_MIG_MIGRATED
&&
254 qp
->s_mig_state
== IB_MIG_ARMED
) {
255 qp
->s_flags
|= RVT_S_AHG_CLEAR
;
256 priv
->s_sc
= ah_to_sc(ibqp
->device
, &qp
->remote_ah_attr
);
257 priv
->s_sde
= qp_to_sdma_engine(qp
, priv
->s_sc
);
258 priv
->s_sendcontext
= qp_to_send_context(qp
, priv
->s_sc
);
263 * hfi1_check_send_wqe - validate wqe
265 * @wqe - The built wqe
267 * validate wqe. This is called
268 * prior to inserting the wqe into
269 * the ring but after the wqe has been
272 * Returns 0 on success, -EINVAL on failure
275 int hfi1_check_send_wqe(struct rvt_qp
*qp
,
276 struct rvt_swqe
*wqe
)
278 struct hfi1_ibport
*ibp
= to_iport(qp
->ibqp
.device
, qp
->port_num
);
281 switch (qp
->ibqp
.qp_type
) {
284 if (wqe
->length
> 0x80000000U
)
288 ah
= ibah_to_rvtah(wqe
->ud_wr
.ah
);
289 if (wqe
->length
> (1 << ah
->log_pmtu
))
294 ah
= ibah_to_rvtah(wqe
->ud_wr
.ah
);
295 if (wqe
->length
> (1 << ah
->log_pmtu
))
297 if (ibp
->sl_to_sc
[rdma_ah_get_sl(&ah
->attr
)] == 0xf)
302 return wqe
->length
<= piothreshold
;
306 * _hfi1_schedule_send - schedule progress
309 * This schedules qp progress w/o regard to the s_flags.
311 * It is only used in the post send, which doesn't hold
314 void _hfi1_schedule_send(struct rvt_qp
*qp
)
316 struct hfi1_qp_priv
*priv
= qp
->priv
;
317 struct hfi1_ibport
*ibp
=
318 to_iport(qp
->ibqp
.device
, qp
->port_num
);
319 struct hfi1_pportdata
*ppd
= ppd_from_ibp(ibp
);
320 struct hfi1_devdata
*dd
= dd_from_ibdev(qp
->ibqp
.device
);
322 iowait_schedule(&priv
->s_iowait
, ppd
->hfi1_wq
,
325 cpumask_first(cpumask_of_node(dd
->node
)));
328 static void qp_pio_drain(struct rvt_qp
*qp
)
330 struct hfi1_ibdev
*dev
;
331 struct hfi1_qp_priv
*priv
= qp
->priv
;
333 if (!priv
->s_sendcontext
)
335 dev
= to_idev(qp
->ibqp
.device
);
336 while (iowait_pio_pending(&priv
->s_iowait
)) {
337 write_seqlock_irq(&dev
->iowait_lock
);
338 hfi1_sc_wantpiobuf_intr(priv
->s_sendcontext
, 1);
339 write_sequnlock_irq(&dev
->iowait_lock
);
340 iowait_pio_drain(&priv
->s_iowait
);
341 write_seqlock_irq(&dev
->iowait_lock
);
342 hfi1_sc_wantpiobuf_intr(priv
->s_sendcontext
, 0);
343 write_sequnlock_irq(&dev
->iowait_lock
);
348 * hfi1_schedule_send - schedule progress
351 * This schedules qp progress and caller should hold
354 void hfi1_schedule_send(struct rvt_qp
*qp
)
356 lockdep_assert_held(&qp
->s_lock
);
357 if (hfi1_send_ok(qp
))
358 _hfi1_schedule_send(qp
);
361 void hfi1_qp_wakeup(struct rvt_qp
*qp
, u32 flag
)
365 spin_lock_irqsave(&qp
->s_lock
, flags
);
366 if (qp
->s_flags
& flag
) {
367 qp
->s_flags
&= ~flag
;
368 trace_hfi1_qpwakeup(qp
, flag
);
369 hfi1_schedule_send(qp
);
371 spin_unlock_irqrestore(&qp
->s_lock
, flags
);
372 /* Notify hfi1_destroy_qp() if it is waiting. */
376 static int iowait_sleep(
377 struct sdma_engine
*sde
,
379 struct sdma_txreq
*stx
,
382 struct verbs_txreq
*tx
= container_of(stx
, struct verbs_txreq
, txreq
);
384 struct hfi1_qp_priv
*priv
;
387 struct hfi1_ibdev
*dev
;
392 spin_lock_irqsave(&qp
->s_lock
, flags
);
393 if (ib_rvt_state_ops
[qp
->state
] & RVT_PROCESS_RECV_OK
) {
395 * If we couldn't queue the DMA request, save the info
396 * and try again later rather than destroying the
397 * buffer and undoing the side effects of the copy.
399 /* Make a common routine? */
400 dev
= &sde
->dd
->verbs_dev
;
401 list_add_tail(&stx
->list
, &wait
->tx_head
);
402 write_seqlock(&dev
->iowait_lock
);
403 if (sdma_progress(sde
, seq
, stx
))
405 if (list_empty(&priv
->s_iowait
.list
)) {
406 struct hfi1_ibport
*ibp
=
407 to_iport(qp
->ibqp
.device
, qp
->port_num
);
409 ibp
->rvp
.n_dmawait
++;
410 qp
->s_flags
|= RVT_S_WAIT_DMA_DESC
;
411 list_add_tail(&priv
->s_iowait
.list
, &sde
->dmawait
);
412 priv
->s_iowait
.lock
= &dev
->iowait_lock
;
413 trace_hfi1_qpsleep(qp
, RVT_S_WAIT_DMA_DESC
);
416 write_sequnlock(&dev
->iowait_lock
);
417 qp
->s_flags
&= ~RVT_S_BUSY
;
418 spin_unlock_irqrestore(&qp
->s_lock
, flags
);
421 spin_unlock_irqrestore(&qp
->s_lock
, flags
);
426 write_sequnlock(&dev
->iowait_lock
);
427 spin_unlock_irqrestore(&qp
->s_lock
, flags
);
428 list_del_init(&stx
->list
);
432 static void iowait_wakeup(struct iowait
*wait
, int reason
)
434 struct rvt_qp
*qp
= iowait_to_qp(wait
);
436 WARN_ON(reason
!= SDMA_AVAIL_REASON
);
437 hfi1_qp_wakeup(qp
, RVT_S_WAIT_DMA_DESC
);
440 static void iowait_sdma_drained(struct iowait
*wait
)
442 struct rvt_qp
*qp
= iowait_to_qp(wait
);
446 * This happens when the send engine notes
447 * a QP in the error state and cannot
448 * do the flush work until that QP's
449 * sdma work has finished.
451 spin_lock_irqsave(&qp
->s_lock
, flags
);
452 if (qp
->s_flags
& RVT_S_WAIT_DMA
) {
453 qp
->s_flags
&= ~RVT_S_WAIT_DMA
;
454 hfi1_schedule_send(qp
);
456 spin_unlock_irqrestore(&qp
->s_lock
, flags
);
461 * qp_to_sdma_engine - map a qp to a send engine
466 * A send engine for the qp or NULL for SMI type qp.
468 struct sdma_engine
*qp_to_sdma_engine(struct rvt_qp
*qp
, u8 sc5
)
470 struct hfi1_devdata
*dd
= dd_from_ibdev(qp
->ibqp
.device
);
471 struct sdma_engine
*sde
;
473 if (!(dd
->flags
& HFI1_HAS_SEND_DMA
))
475 switch (qp
->ibqp
.qp_type
) {
481 sde
= sdma_select_engine_sc(dd
, qp
->ibqp
.qp_num
>> dd
->qos_shift
, sc5
);
486 * qp_to_send_context - map a qp to a send context
491 * A send context for the qp
493 struct send_context
*qp_to_send_context(struct rvt_qp
*qp
, u8 sc5
)
495 struct hfi1_devdata
*dd
= dd_from_ibdev(qp
->ibqp
.device
);
497 switch (qp
->ibqp
.qp_type
) {
499 /* SMA packets to VL15 */
500 return dd
->vld
[15].sc
;
505 return pio_select_send_context_sc(dd
, qp
->ibqp
.qp_num
>> dd
->qos_shift
,
510 struct hfi1_ibdev
*dev
;
516 struct qp_iter
*qp_iter_init(struct hfi1_ibdev
*dev
)
518 struct qp_iter
*iter
;
520 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
525 iter
->specials
= dev
->rdi
.ibdev
.phys_port_cnt
* 2;
530 int qp_iter_next(struct qp_iter
*iter
)
532 struct hfi1_ibdev
*dev
= iter
->dev
;
535 struct rvt_qp
*pqp
= iter
->qp
;
539 * The approach is to consider the special qps
540 * as an additional table entries before the
541 * real hash table. Since the qp code sets
542 * the qp->next hash link to NULL, this works just fine.
544 * iter->specials is 2 * # ports
546 * n = 0..iter->specials is the special qp indices
548 * n = iter->specials..dev->rdi.qp_dev->qp_table_size+iter->specials are
549 * the potential hash bucket entries
552 for (; n
< dev
->rdi
.qp_dev
->qp_table_size
+ iter
->specials
; n
++) {
554 qp
= rcu_dereference(pqp
->next
);
556 if (n
< iter
->specials
) {
557 struct hfi1_pportdata
*ppd
;
558 struct hfi1_ibport
*ibp
;
561 pidx
= n
% dev
->rdi
.ibdev
.phys_port_cnt
;
562 ppd
= &dd_from_dev(dev
)->pport
[pidx
];
563 ibp
= &ppd
->ibport_data
;
566 qp
= rcu_dereference(ibp
->rvp
.qp
[0]);
568 qp
= rcu_dereference(ibp
->rvp
.qp
[1]);
570 qp
= rcu_dereference(
571 dev
->rdi
.qp_dev
->qp_table
[
572 (n
- iter
->specials
)]);
585 static const char * const qp_type_str
[] = {
586 "SMI", "GSI", "RC", "UC", "UD",
589 static int qp_idle(struct rvt_qp
*qp
)
592 qp
->s_last
== qp
->s_acked
&&
593 qp
->s_acked
== qp
->s_cur
&&
594 qp
->s_cur
== qp
->s_tail
&&
595 qp
->s_tail
== qp
->s_head
;
598 void qp_iter_print(struct seq_file
*s
, struct qp_iter
*iter
)
600 struct rvt_swqe
*wqe
;
601 struct rvt_qp
*qp
= iter
->qp
;
602 struct hfi1_qp_priv
*priv
= qp
->priv
;
603 struct sdma_engine
*sde
;
604 struct send_context
*send_context
;
606 sde
= qp_to_sdma_engine(qp
, priv
->s_sc
);
607 wqe
= rvt_get_swqe_ptr(qp
, qp
->s_last
);
608 send_context
= qp_to_send_context(qp
, priv
->s_sc
);
610 "N %d %s QP %x R %u %s %u %u %u f=%x %u %u %u %u %u %u SPSN %x %x %x %x %x RPSN %x (%u %u %u %u %u %u %u) RQP %x LID %x SL %u MTU %u %u %u %u %u SDE %p,%u SC %p,%u SCQ %u %u PID %d\n",
612 qp_idle(qp
) ? "I" : "B",
614 atomic_read(&qp
->refcount
),
615 qp_type_str
[qp
->ibqp
.qp_type
],
617 wqe
? wqe
->wr
.opcode
: 0,
620 iowait_sdma_pending(&priv
->s_iowait
),
621 iowait_pio_pending(&priv
->s_iowait
),
622 !list_empty(&priv
->s_iowait
.list
),
627 qp
->s_psn
, qp
->s_next_psn
,
628 qp
->s_sending_psn
, qp
->s_sending_hpsn
,
630 qp
->s_last
, qp
->s_acked
, qp
->s_cur
,
631 qp
->s_tail
, qp
->s_head
, qp
->s_size
,
634 rdma_ah_get_dlid(&qp
->remote_ah_attr
),
635 rdma_ah_get_sl(&qp
->remote_ah_attr
),
642 sde
? sde
->this_idx
: 0,
644 send_context
? send_context
->sw_index
: 0,
645 ibcq_to_rvtcq(qp
->ibqp
.send_cq
)->queue
->head
,
646 ibcq_to_rvtcq(qp
->ibqp
.send_cq
)->queue
->tail
,
650 void *qp_priv_alloc(struct rvt_dev_info
*rdi
, struct rvt_qp
*qp
,
653 struct hfi1_qp_priv
*priv
;
655 priv
= kzalloc_node(sizeof(*priv
), gfp
, rdi
->dparms
.node
);
657 return ERR_PTR(-ENOMEM
);
661 priv
->s_ahg
= kzalloc_node(sizeof(*priv
->s_ahg
), gfp
,
665 return ERR_PTR(-ENOMEM
);
673 iowait_sdma_drained
);
677 void qp_priv_free(struct rvt_dev_info
*rdi
, struct rvt_qp
*qp
)
679 struct hfi1_qp_priv
*priv
= qp
->priv
;
685 unsigned free_all_qps(struct rvt_dev_info
*rdi
)
687 struct hfi1_ibdev
*verbs_dev
= container_of(rdi
,
690 struct hfi1_devdata
*dd
= container_of(verbs_dev
,
694 unsigned qp_inuse
= 0;
696 for (n
= 0; n
< dd
->num_pports
; n
++) {
697 struct hfi1_ibport
*ibp
= &dd
->pport
[n
].ibport_data
;
700 if (rcu_dereference(ibp
->rvp
.qp
[0]))
702 if (rcu_dereference(ibp
->rvp
.qp
[1]))
710 void flush_qp_waiters(struct rvt_qp
*qp
)
712 lockdep_assert_held(&qp
->s_lock
);
716 void stop_send_queue(struct rvt_qp
*qp
)
718 struct hfi1_qp_priv
*priv
= qp
->priv
;
720 cancel_work_sync(&priv
->s_iowait
.iowork
);
723 void quiesce_qp(struct rvt_qp
*qp
)
725 struct hfi1_qp_priv
*priv
= qp
->priv
;
727 iowait_sdma_drain(&priv
->s_iowait
);
732 void notify_qp_reset(struct rvt_qp
*qp
)
739 * Switch to alternate path.
740 * The QP s_lock should be held and interrupts disabled.
742 void hfi1_migrate_qp(struct rvt_qp
*qp
)
744 struct hfi1_qp_priv
*priv
= qp
->priv
;
747 qp
->s_mig_state
= IB_MIG_MIGRATED
;
748 qp
->remote_ah_attr
= qp
->alt_ah_attr
;
749 qp
->port_num
= rdma_ah_get_port_num(&qp
->alt_ah_attr
);
750 qp
->s_pkey_index
= qp
->s_alt_pkey_index
;
751 qp
->s_flags
|= RVT_S_AHG_CLEAR
;
752 priv
->s_sc
= ah_to_sc(qp
->ibqp
.device
, &qp
->remote_ah_attr
);
753 priv
->s_sde
= qp_to_sdma_engine(qp
, priv
->s_sc
);
755 ev
.device
= qp
->ibqp
.device
;
756 ev
.element
.qp
= &qp
->ibqp
;
757 ev
.event
= IB_EVENT_PATH_MIG
;
758 qp
->ibqp
.event_handler(&ev
, qp
->ibqp
.qp_context
);
761 int mtu_to_path_mtu(u32 mtu
)
763 return mtu_to_enum(mtu
, OPA_MTU_8192
);
766 u32
mtu_from_qp(struct rvt_dev_info
*rdi
, struct rvt_qp
*qp
, u32 pmtu
)
769 struct hfi1_ibdev
*verbs_dev
= container_of(rdi
,
772 struct hfi1_devdata
*dd
= container_of(verbs_dev
,
775 struct hfi1_ibport
*ibp
;
778 ibp
= &dd
->pport
[qp
->port_num
- 1].ibport_data
;
779 sc
= ibp
->sl_to_sc
[rdma_ah_get_sl(&qp
->remote_ah_attr
)];
780 vl
= sc_to_vlt(dd
, sc
);
782 mtu
= verbs_mtu_enum_to_int(qp
->ibqp
.device
, pmtu
);
783 if (vl
< PER_VL_SEND_CONTEXTS
)
784 mtu
= min_t(u32
, mtu
, dd
->vld
[vl
].mtu
);
788 int get_pmtu_from_attr(struct rvt_dev_info
*rdi
, struct rvt_qp
*qp
,
789 struct ib_qp_attr
*attr
)
791 int mtu
, pidx
= qp
->port_num
- 1;
792 struct hfi1_ibdev
*verbs_dev
= container_of(rdi
,
795 struct hfi1_devdata
*dd
= container_of(verbs_dev
,
798 mtu
= verbs_mtu_enum_to_int(qp
->ibqp
.device
, attr
->path_mtu
);
800 return -1; /* values less than 0 are error */
802 if (mtu
> dd
->pport
[pidx
].ibmtu
)
803 return mtu_to_enum(dd
->pport
[pidx
].ibmtu
, IB_MTU_2048
);
805 return attr
->path_mtu
;
808 void notify_error_qp(struct rvt_qp
*qp
)
810 struct hfi1_qp_priv
*priv
= qp
->priv
;
811 seqlock_t
*lock
= priv
->s_iowait
.lock
;
815 if (!list_empty(&priv
->s_iowait
.list
) &&
816 !(qp
->s_flags
& RVT_S_BUSY
)) {
817 qp
->s_flags
&= ~RVT_S_ANY_WAIT_IO
;
818 list_del_init(&priv
->s_iowait
.list
);
819 priv
->s_iowait
.lock
= NULL
;
822 write_sequnlock(lock
);
825 if (!(qp
->s_flags
& RVT_S_BUSY
)) {
828 rvt_put_mr(qp
->s_rdma_mr
);
829 qp
->s_rdma_mr
= NULL
;
836 * hfi1_error_port_qps - put a port's RC/UC qps into error state
838 * @sl: the service level.
840 * This function places all RC/UC qps with a given service level into error
841 * state. It is generally called to force upper lay apps to abandon stale qps
842 * after an sl->sc mapping change.
844 void hfi1_error_port_qps(struct hfi1_ibport
*ibp
, u8 sl
)
846 struct rvt_qp
*qp
= NULL
;
847 struct hfi1_pportdata
*ppd
= ppd_from_ibp(ibp
);
848 struct hfi1_ibdev
*dev
= &ppd
->dd
->verbs_dev
;
855 /* Deal only with RC/UC qps that use the given SL. */
856 for (n
= 0; n
< dev
->rdi
.qp_dev
->qp_table_size
; n
++) {
857 for (qp
= rcu_dereference(dev
->rdi
.qp_dev
->qp_table
[n
]); qp
;
858 qp
= rcu_dereference(qp
->next
)) {
859 if (qp
->port_num
== ppd
->port
&&
860 (qp
->ibqp
.qp_type
== IB_QPT_UC
||
861 qp
->ibqp
.qp_type
== IB_QPT_RC
) &&
862 rdma_ah_get_sl(&qp
->remote_ah_attr
) == sl
&&
863 (ib_rvt_state_ops
[qp
->state
] &
865 spin_lock_irq(&qp
->r_lock
);
866 spin_lock(&qp
->s_hlock
);
867 spin_lock(&qp
->s_lock
);
868 lastwqe
= rvt_error_qp(qp
,
870 spin_unlock(&qp
->s_lock
);
871 spin_unlock(&qp
->s_hlock
);
872 spin_unlock_irq(&qp
->r_lock
);
874 ev
.device
= qp
->ibqp
.device
;
875 ev
.element
.qp
= &qp
->ibqp
;
877 IB_EVENT_QP_LAST_WQE_REACHED
;
878 qp
->ibqp
.event_handler(&ev
,
879 qp
->ibqp
.qp_context
);