IB/ipath: merge ipath_core and ib_ipath drivers
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / drivers / infiniband / hw / ipath / ipath_verbs.c
CommitLineData
6522108f 1/*
759d5768 2 * Copyright (c) 2006 QLogic, Inc. All rights reserved.
6522108f
BS
3 * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
4 *
5 * This software is available to you under a choice of one of two
6 * licenses. You may choose to be licensed under the terms of the GNU
7 * General Public License (GPL) Version 2, available from the file
8 * COPYING in the main directory of this source tree, or the
9 * OpenIB.org BSD license below:
10 *
11 * Redistribution and use in source and binary forms, with or
12 * without modification, are permitted provided that the following
13 * conditions are met:
14 *
15 * - Redistributions of source code must retain the above
16 * copyright notice, this list of conditions and the following
17 * disclaimer.
18 *
19 * - Redistributions in binary form must reproduce the above
20 * copyright notice, this list of conditions and the following
21 * disclaimer in the documentation and/or other materials
22 * provided with the distribution.
23 *
24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31 * SOFTWARE.
32 */
33
34#include <rdma/ib_mad.h>
35#include <rdma/ib_user_verbs.h>
36#include <linux/utsname.h>
37
38#include "ipath_kernel.h"
39#include "ipath_verbs.h"
27b678dd 40#include "ipath_common.h"
6522108f
BS
41
42/* Not static, because we don't want the compiler removing it */
43const char ipath_verbs_version[] = "ipath_verbs " IPATH_IDSTR;
44
ac2ae4c9 45static unsigned int ib_ipath_qp_table_size = 251;
6522108f
BS
46module_param_named(qp_table_size, ib_ipath_qp_table_size, uint, S_IRUGO);
47MODULE_PARM_DESC(qp_table_size, "QP table size");
48
49unsigned int ib_ipath_lkey_table_size = 12;
50module_param_named(lkey_table_size, ib_ipath_lkey_table_size, uint,
51 S_IRUGO);
52MODULE_PARM_DESC(lkey_table_size,
53 "LKEY table size in bits (2^n, 1 <= n <= 23)");
54
55unsigned int ib_ipath_debug; /* debug mask */
56module_param_named(debug, ib_ipath_debug, uint, S_IWUSR | S_IRUGO);
57MODULE_PARM_DESC(debug, "Verbs debug mask");
58
fe62546a
BS
59static unsigned int ib_ipath_max_pds = 0xFFFF;
60module_param_named(max_pds, ib_ipath_max_pds, uint, S_IWUSR | S_IRUGO);
61MODULE_PARM_DESC(max_pds,
62 "Maximum number of protection domains to support");
63
64static unsigned int ib_ipath_max_ahs = 0xFFFF;
65module_param_named(max_ahs, ib_ipath_max_ahs, uint, S_IWUSR | S_IRUGO);
66MODULE_PARM_DESC(max_ahs, "Maximum number of address handles to support");
67
68unsigned int ib_ipath_max_cqes = 0x2FFFF;
69module_param_named(max_cqes, ib_ipath_max_cqes, uint, S_IWUSR | S_IRUGO);
70MODULE_PARM_DESC(max_cqes,
71 "Maximum number of completion queue entries to support");
72
73unsigned int ib_ipath_max_cqs = 0x1FFFF;
74module_param_named(max_cqs, ib_ipath_max_cqs, uint, S_IWUSR | S_IRUGO);
75MODULE_PARM_DESC(max_cqs, "Maximum number of completion queues to support");
76
77unsigned int ib_ipath_max_qp_wrs = 0x3FFF;
78module_param_named(max_qp_wrs, ib_ipath_max_qp_wrs, uint,
79 S_IWUSR | S_IRUGO);
80MODULE_PARM_DESC(max_qp_wrs, "Maximum number of QP WRs to support");
81
82unsigned int ib_ipath_max_sges = 0x60;
83module_param_named(max_sges, ib_ipath_max_sges, uint, S_IWUSR | S_IRUGO);
84MODULE_PARM_DESC(max_sges, "Maximum number of SGEs to support");
85
86unsigned int ib_ipath_max_mcast_grps = 16384;
87module_param_named(max_mcast_grps, ib_ipath_max_mcast_grps, uint,
88 S_IWUSR | S_IRUGO);
89MODULE_PARM_DESC(max_mcast_grps,
90 "Maximum number of multicast groups to support");
91
92unsigned int ib_ipath_max_mcast_qp_attached = 16;
93module_param_named(max_mcast_qp_attached, ib_ipath_max_mcast_qp_attached,
94 uint, S_IWUSR | S_IRUGO);
95MODULE_PARM_DESC(max_mcast_qp_attached,
96 "Maximum number of attached QPs to support");
97
98unsigned int ib_ipath_max_srqs = 1024;
99module_param_named(max_srqs, ib_ipath_max_srqs, uint, S_IWUSR | S_IRUGO);
100MODULE_PARM_DESC(max_srqs, "Maximum number of SRQs to support");
101
102unsigned int ib_ipath_max_srq_sges = 128;
103module_param_named(max_srq_sges, ib_ipath_max_srq_sges,
104 uint, S_IWUSR | S_IRUGO);
105MODULE_PARM_DESC(max_srq_sges, "Maximum number of SRQ SGEs to support");
106
107unsigned int ib_ipath_max_srq_wrs = 0x1FFFF;
108module_param_named(max_srq_wrs, ib_ipath_max_srq_wrs,
109 uint, S_IWUSR | S_IRUGO);
110MODULE_PARM_DESC(max_srq_wrs, "Maximum number of SRQ WRs support");
111
6522108f 112MODULE_LICENSE("GPL");
759d5768
BS
113MODULE_AUTHOR("QLogic <support@pathscale.com>");
114MODULE_DESCRIPTION("QLogic InfiniPath driver");
6522108f
BS
115
116const int ib_ipath_state_ops[IB_QPS_ERR + 1] = {
117 [IB_QPS_RESET] = 0,
118 [IB_QPS_INIT] = IPATH_POST_RECV_OK,
119 [IB_QPS_RTR] = IPATH_POST_RECV_OK | IPATH_PROCESS_RECV_OK,
120 [IB_QPS_RTS] = IPATH_POST_RECV_OK | IPATH_PROCESS_RECV_OK |
121 IPATH_POST_SEND_OK | IPATH_PROCESS_SEND_OK,
122 [IB_QPS_SQD] = IPATH_POST_RECV_OK | IPATH_PROCESS_RECV_OK |
123 IPATH_POST_SEND_OK,
124 [IB_QPS_SQE] = IPATH_POST_RECV_OK | IPATH_PROCESS_RECV_OK,
125 [IB_QPS_ERR] = 0,
126};
127
128/*
129 * Translate ib_wr_opcode into ib_wc_opcode.
130 */
131const enum ib_wc_opcode ib_ipath_wc_opcode[] = {
132 [IB_WR_RDMA_WRITE] = IB_WC_RDMA_WRITE,
133 [IB_WR_RDMA_WRITE_WITH_IMM] = IB_WC_RDMA_WRITE,
134 [IB_WR_SEND] = IB_WC_SEND,
135 [IB_WR_SEND_WITH_IMM] = IB_WC_SEND,
136 [IB_WR_RDMA_READ] = IB_WC_RDMA_READ,
137 [IB_WR_ATOMIC_CMP_AND_SWP] = IB_WC_COMP_SWAP,
138 [IB_WR_ATOMIC_FETCH_AND_ADD] = IB_WC_FETCH_ADD
139};
140
141/*
142 * System image GUID.
143 */
ac2ae4c9 144static __be64 sys_image_guid;
6522108f
BS
145
146/**
147 * ipath_copy_sge - copy data to SGE memory
148 * @ss: the SGE state
149 * @data: the data to copy
150 * @length: the length of the data
151 */
152void ipath_copy_sge(struct ipath_sge_state *ss, void *data, u32 length)
153{
154 struct ipath_sge *sge = &ss->sge;
155
156 while (length) {
157 u32 len = sge->length;
158
159 BUG_ON(len == 0);
160 if (len > length)
161 len = length;
162 memcpy(sge->vaddr, data, len);
163 sge->vaddr += len;
164 sge->length -= len;
165 sge->sge_length -= len;
166 if (sge->sge_length == 0) {
167 if (--ss->num_sge)
168 *sge = *ss->sg_list++;
169 } else if (sge->length == 0 && sge->mr != NULL) {
170 if (++sge->n >= IPATH_SEGSZ) {
171 if (++sge->m >= sge->mr->mapsz)
172 break;
173 sge->n = 0;
174 }
175 sge->vaddr =
176 sge->mr->map[sge->m]->segs[sge->n].vaddr;
177 sge->length =
178 sge->mr->map[sge->m]->segs[sge->n].length;
179 }
180 data += len;
181 length -= len;
182 }
183}
184
185/**
186 * ipath_skip_sge - skip over SGE memory - XXX almost dup of prev func
187 * @ss: the SGE state
188 * @length: the number of bytes to skip
189 */
190void ipath_skip_sge(struct ipath_sge_state *ss, u32 length)
191{
192 struct ipath_sge *sge = &ss->sge;
193
6522108f
BS
194 while (length) {
195 u32 len = sge->length;
196
197 BUG_ON(len == 0);
198 if (len > length)
199 len = length;
200 sge->vaddr += len;
201 sge->length -= len;
202 sge->sge_length -= len;
203 if (sge->sge_length == 0) {
204 if (--ss->num_sge)
205 *sge = *ss->sg_list++;
206 } else if (sge->length == 0 && sge->mr != NULL) {
207 if (++sge->n >= IPATH_SEGSZ) {
208 if (++sge->m >= sge->mr->mapsz)
209 break;
210 sge->n = 0;
211 }
212 sge->vaddr =
213 sge->mr->map[sge->m]->segs[sge->n].vaddr;
214 sge->length =
215 sge->mr->map[sge->m]->segs[sge->n].length;
216 }
217 length -= len;
218 }
219}
220
221/**
222 * ipath_post_send - post a send on a QP
223 * @ibqp: the QP to post the send on
224 * @wr: the list of work requests to post
225 * @bad_wr: the first bad WR is put here
226 *
227 * This may be called from interrupt context.
228 */
229static int ipath_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
230 struct ib_send_wr **bad_wr)
231{
232 struct ipath_qp *qp = to_iqp(ibqp);
233 int err = 0;
234
235 /* Check that state is OK to post send. */
236 if (!(ib_ipath_state_ops[qp->state] & IPATH_POST_SEND_OK)) {
237 *bad_wr = wr;
238 err = -EINVAL;
239 goto bail;
240 }
241
242 for (; wr; wr = wr->next) {
243 switch (qp->ibqp.qp_type) {
244 case IB_QPT_UC:
245 case IB_QPT_RC:
ddd4bb22 246 err = ipath_post_ruc_send(qp, wr);
6522108f
BS
247 break;
248
249 case IB_QPT_SMI:
250 case IB_QPT_GSI:
251 case IB_QPT_UD:
252 err = ipath_post_ud_send(qp, wr);
253 break;
254
255 default:
256 err = -EINVAL;
257 }
258 if (err) {
259 *bad_wr = wr;
260 break;
261 }
262 }
263
264bail:
265 return err;
266}
267
268/**
269 * ipath_post_receive - post a receive on a QP
270 * @ibqp: the QP to post the receive on
271 * @wr: the WR to post
272 * @bad_wr: the first bad WR is put here
273 *
274 * This may be called from interrupt context.
275 */
276static int ipath_post_receive(struct ib_qp *ibqp, struct ib_recv_wr *wr,
277 struct ib_recv_wr **bad_wr)
278{
279 struct ipath_qp *qp = to_iqp(ibqp);
373d9915 280 struct ipath_rwq *wq = qp->r_rq.wq;
6522108f
BS
281 unsigned long flags;
282 int ret;
283
284 /* Check that state is OK to post receive. */
373d9915 285 if (!(ib_ipath_state_ops[qp->state] & IPATH_POST_RECV_OK) || !wq) {
6522108f
BS
286 *bad_wr = wr;
287 ret = -EINVAL;
288 goto bail;
289 }
290
291 for (; wr; wr = wr->next) {
292 struct ipath_rwqe *wqe;
293 u32 next;
373d9915 294 int i;
6522108f 295
373d9915 296 if ((unsigned) wr->num_sge > qp->r_rq.max_sge) {
6522108f
BS
297 *bad_wr = wr;
298 ret = -ENOMEM;
299 goto bail;
300 }
301
302 spin_lock_irqsave(&qp->r_rq.lock, flags);
373d9915 303 next = wq->head + 1;
6522108f
BS
304 if (next >= qp->r_rq.size)
305 next = 0;
373d9915 306 if (next == wq->tail) {
6522108f
BS
307 spin_unlock_irqrestore(&qp->r_rq.lock, flags);
308 *bad_wr = wr;
309 ret = -ENOMEM;
310 goto bail;
311 }
312
373d9915 313 wqe = get_rwqe_ptr(&qp->r_rq, wq->head);
6522108f 314 wqe->wr_id = wr->wr_id;
373d9915
RC
315 wqe->num_sge = wr->num_sge;
316 for (i = 0; i < wr->num_sge; i++)
317 wqe->sg_list[i] = wr->sg_list[i];
318 wq->head = next;
6522108f
BS
319 spin_unlock_irqrestore(&qp->r_rq.lock, flags);
320 }
321 ret = 0;
322
323bail:
324 return ret;
325}
326
327/**
328 * ipath_qp_rcv - processing an incoming packet on a QP
329 * @dev: the device the packet came on
330 * @hdr: the packet header
331 * @has_grh: true if the packet has a GRH
332 * @data: the packet data
333 * @tlen: the packet length
334 * @qp: the QP the packet came on
335 *
336 * This is called from ipath_ib_rcv() to process an incoming packet
337 * for the given QP.
338 * Called at interrupt level.
339 */
340static void ipath_qp_rcv(struct ipath_ibdev *dev,
341 struct ipath_ib_header *hdr, int has_grh,
342 void *data, u32 tlen, struct ipath_qp *qp)
343{
344 /* Check for valid receive state. */
345 if (!(ib_ipath_state_ops[qp->state] & IPATH_PROCESS_RECV_OK)) {
346 dev->n_pkt_drops++;
347 return;
348 }
349
350 switch (qp->ibqp.qp_type) {
351 case IB_QPT_SMI:
352 case IB_QPT_GSI:
353 case IB_QPT_UD:
354 ipath_ud_rcv(dev, hdr, has_grh, data, tlen, qp);
355 break;
356
357 case IB_QPT_RC:
358 ipath_rc_rcv(dev, hdr, has_grh, data, tlen, qp);
359 break;
360
361 case IB_QPT_UC:
362 ipath_uc_rcv(dev, hdr, has_grh, data, tlen, qp);
363 break;
364
365 default:
366 break;
367 }
368}
369
370/**
b1c1b6a3 371 * ipath_ib_rcv - process an incoming packet
6522108f
BS
372 * @arg: the device pointer
373 * @rhdr: the header of the packet
374 * @data: the packet data
375 * @tlen: the packet length
376 *
377 * This is called from ipath_kreceive() to process an incoming packet at
378 * interrupt level. Tlen is the length of the header + data + CRC in bytes.
379 */
b1c1b6a3
BS
380void ipath_ib_rcv(struct ipath_ibdev *dev, void *rhdr, void *data,
381 u32 tlen)
6522108f 382{
6522108f
BS
383 struct ipath_ib_header *hdr = rhdr;
384 struct ipath_other_headers *ohdr;
385 struct ipath_qp *qp;
386 u32 qp_num;
387 int lnh;
388 u8 opcode;
389 u16 lid;
390
391 if (unlikely(dev == NULL))
392 goto bail;
393
394 if (unlikely(tlen < 24)) { /* LRH+BTH+CRC */
395 dev->rcv_errors++;
396 goto bail;
397 }
398
399 /* Check for a valid destination LID (see ch. 7.11.1). */
400 lid = be16_to_cpu(hdr->lrh[1]);
27b678dd 401 if (lid < IPATH_MULTICAST_LID_BASE) {
6522108f
BS
402 lid &= ~((1 << (dev->mkeyprot_resv_lmc & 7)) - 1);
403 if (unlikely(lid != ipath_layer_get_lid(dev->dd))) {
404 dev->rcv_errors++;
405 goto bail;
406 }
407 }
408
409 /* Check for GRH */
410 lnh = be16_to_cpu(hdr->lrh[0]) & 3;
27b678dd 411 if (lnh == IPATH_LRH_BTH)
6522108f 412 ohdr = &hdr->u.oth;
27b678dd 413 else if (lnh == IPATH_LRH_GRH)
6522108f
BS
414 ohdr = &hdr->u.l.oth;
415 else {
416 dev->rcv_errors++;
417 goto bail;
418 }
419
420 opcode = be32_to_cpu(ohdr->bth[0]) >> 24;
421 dev->opstats[opcode].n_bytes += tlen;
422 dev->opstats[opcode].n_packets++;
423
424 /* Get the destination QP number. */
27b678dd
BS
425 qp_num = be32_to_cpu(ohdr->bth[1]) & IPATH_QPN_MASK;
426 if (qp_num == IPATH_MULTICAST_QPN) {
6522108f
BS
427 struct ipath_mcast *mcast;
428 struct ipath_mcast_qp *p;
429
430 mcast = ipath_mcast_find(&hdr->u.l.grh.dgid);
431 if (mcast == NULL) {
432 dev->n_pkt_drops++;
433 goto bail;
434 }
435 dev->n_multicast_rcv++;
436 list_for_each_entry_rcu(p, &mcast->qp_list, list)
27b678dd 437 ipath_qp_rcv(dev, hdr, lnh == IPATH_LRH_GRH, data,
6522108f
BS
438 tlen, p->qp);
439 /*
440 * Notify ipath_multicast_detach() if it is waiting for us
441 * to finish.
442 */
443 if (atomic_dec_return(&mcast->refcount) <= 1)
444 wake_up(&mcast->wait);
445 } else {
446 qp = ipath_lookup_qpn(&dev->qp_table, qp_num);
447 if (qp) {
448 dev->n_unicast_rcv++;
27b678dd 449 ipath_qp_rcv(dev, hdr, lnh == IPATH_LRH_GRH, data,
6522108f
BS
450 tlen, qp);
451 /*
452 * Notify ipath_destroy_qp() if it is waiting
453 * for us to finish.
454 */
455 if (atomic_dec_and_test(&qp->refcount))
456 wake_up(&qp->wait);
457 } else
458 dev->n_pkt_drops++;
459 }
460
461bail:;
462}
463
464/**
465 * ipath_ib_timer - verbs timer
466 * @arg: the device pointer
467 *
468 * This is called from ipath_do_rcv_timer() at interrupt level to check for
469 * QPs which need retransmits and to collect performance numbers.
470 */
b1c1b6a3 471void ipath_ib_timer(struct ipath_ibdev *dev)
6522108f 472{
6522108f 473 struct ipath_qp *resend = NULL;
6522108f
BS
474 struct list_head *last;
475 struct ipath_qp *qp;
476 unsigned long flags;
477
478 if (dev == NULL)
479 return;
480
481 spin_lock_irqsave(&dev->pending_lock, flags);
482 /* Start filling the next pending queue. */
483 if (++dev->pending_index >= ARRAY_SIZE(dev->pending))
484 dev->pending_index = 0;
485 /* Save any requests still in the new queue, they have timed out. */
486 last = &dev->pending[dev->pending_index];
487 while (!list_empty(last)) {
488 qp = list_entry(last->next, struct ipath_qp, timerwait);
94b8d9f9 489 list_del_init(&qp->timerwait);
9b2017f1
BS
490 qp->timer_next = resend;
491 resend = qp;
492 atomic_inc(&qp->refcount);
6522108f
BS
493 }
494 last = &dev->rnrwait;
495 if (!list_empty(last)) {
496 qp = list_entry(last->next, struct ipath_qp, timerwait);
497 if (--qp->s_rnr_timeout == 0) {
498 do {
94b8d9f9 499 list_del_init(&qp->timerwait);
9b2017f1 500 tasklet_hi_schedule(&qp->s_task);
6522108f
BS
501 if (list_empty(last))
502 break;
503 qp = list_entry(last->next, struct ipath_qp,
504 timerwait);
505 } while (qp->s_rnr_timeout == 0);
506 }
507 }
508 /*
509 * We should only be in the started state if pma_sample_start != 0
510 */
511 if (dev->pma_sample_status == IB_PMA_SAMPLE_STATUS_STARTED &&
512 --dev->pma_sample_start == 0) {
513 dev->pma_sample_status = IB_PMA_SAMPLE_STATUS_RUNNING;
514 ipath_layer_snapshot_counters(dev->dd, &dev->ipath_sword,
515 &dev->ipath_rword,
516 &dev->ipath_spkts,
517 &dev->ipath_rpkts,
518 &dev->ipath_xmit_wait);
519 }
520 if (dev->pma_sample_status == IB_PMA_SAMPLE_STATUS_RUNNING) {
521 if (dev->pma_sample_interval == 0) {
522 u64 ta, tb, tc, td, te;
523
524 dev->pma_sample_status = IB_PMA_SAMPLE_STATUS_DONE;
525 ipath_layer_snapshot_counters(dev->dd, &ta, &tb,
526 &tc, &td, &te);
527
528 dev->ipath_sword = ta - dev->ipath_sword;
529 dev->ipath_rword = tb - dev->ipath_rword;
530 dev->ipath_spkts = tc - dev->ipath_spkts;
531 dev->ipath_rpkts = td - dev->ipath_rpkts;
532 dev->ipath_xmit_wait = te - dev->ipath_xmit_wait;
533 }
534 else
535 dev->pma_sample_interval--;
536 }
537 spin_unlock_irqrestore(&dev->pending_lock, flags);
538
539 /* XXX What if timer fires again while this is running? */
9b2017f1 540 for (qp = resend; qp != NULL; qp = qp->timer_next) {
6522108f
BS
541 struct ib_wc wc;
542
543 spin_lock_irqsave(&qp->s_lock, flags);
544 if (qp->s_last != qp->s_tail && qp->state == IB_QPS_RTS) {
545 dev->n_timeouts++;
546 ipath_restart_rc(qp, qp->s_last_psn + 1, &wc);
547 }
548 spin_unlock_irqrestore(&qp->s_lock, flags);
549
550 /* Notify ipath_destroy_qp() if it is waiting. */
551 if (atomic_dec_and_test(&qp->refcount))
552 wake_up(&qp->wait);
553 }
6522108f
BS
554}
555
556/**
557 * ipath_ib_piobufavail - callback when a PIO buffer is available
558 * @arg: the device pointer
559 *
560 * This is called from ipath_intr() at interrupt level when a PIO buffer is
561 * available after ipath_verbs_send() returned an error that no buffers were
9b2017f1 562 * available. Return 1 if we consumed all the PIO buffers and we still have
6522108f 563 * QPs waiting for buffers (for now, just do a tasklet_hi_schedule and
9b2017f1 564 * return zero).
6522108f 565 */
b1c1b6a3 566int ipath_ib_piobufavail(struct ipath_ibdev *dev)
6522108f 567{
6522108f
BS
568 struct ipath_qp *qp;
569 unsigned long flags;
570
571 if (dev == NULL)
572 goto bail;
573
574 spin_lock_irqsave(&dev->pending_lock, flags);
575 while (!list_empty(&dev->piowait)) {
576 qp = list_entry(dev->piowait.next, struct ipath_qp,
577 piowait);
94b8d9f9 578 list_del_init(&qp->piowait);
6522108f
BS
579 tasklet_hi_schedule(&qp->s_task);
580 }
581 spin_unlock_irqrestore(&dev->pending_lock, flags);
582
583bail:
9b2017f1 584 return 0;
6522108f
BS
585}
586
587static int ipath_query_device(struct ib_device *ibdev,
588 struct ib_device_attr *props)
589{
590 struct ipath_ibdev *dev = to_idev(ibdev);
6522108f
BS
591
592 memset(props, 0, sizeof(*props));
593
594 props->device_cap_flags = IB_DEVICE_BAD_PKEY_CNTR |
595 IB_DEVICE_BAD_QKEY_CNTR | IB_DEVICE_SHUTDOWN_PORT |
596 IB_DEVICE_SYS_IMAGE_GUID;
c9f79bdc 597 props->page_size_cap = PAGE_SIZE;
e8a88f09
BS
598 props->vendor_id = ipath_layer_get_vendorid(dev->dd);
599 props->vendor_part_id = ipath_layer_get_deviceid(dev->dd);
600 props->hw_ver = ipath_layer_get_pcirev(dev->dd);
6522108f
BS
601
602 props->sys_image_guid = dev->sys_image_guid;
603
604 props->max_mr_size = ~0ull;
fe62546a
BS
605 props->max_qp = dev->qp_table.max;
606 props->max_qp_wr = ib_ipath_max_qp_wrs;
607 props->max_sge = ib_ipath_max_sges;
608 props->max_cq = ib_ipath_max_cqs;
609 props->max_ah = ib_ipath_max_ahs;
610 props->max_cqe = ib_ipath_max_cqes;
611 props->max_mr = dev->lk_table.max;
612 props->max_pd = ib_ipath_max_pds;
6522108f
BS
613 props->max_qp_rd_atom = 1;
614 props->max_qp_init_rd_atom = 1;
615 /* props->max_res_rd_atom */
fe62546a
BS
616 props->max_srq = ib_ipath_max_srqs;
617 props->max_srq_wr = ib_ipath_max_srq_wrs;
618 props->max_srq_sge = ib_ipath_max_srq_sges;
6522108f
BS
619 /* props->local_ca_ack_delay */
620 props->atomic_cap = IB_ATOMIC_HCA;
621 props->max_pkeys = ipath_layer_get_npkeys(dev->dd);
fe62546a
BS
622 props->max_mcast_grp = ib_ipath_max_mcast_grps;
623 props->max_mcast_qp_attach = ib_ipath_max_mcast_qp_attached;
6522108f
BS
624 props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
625 props->max_mcast_grp;
626
627 return 0;
628}
629
630const u8 ipath_cvt_physportstate[16] = {
631 [INFINIPATH_IBCS_LT_STATE_DISABLED] = 3,
632 [INFINIPATH_IBCS_LT_STATE_LINKUP] = 5,
633 [INFINIPATH_IBCS_LT_STATE_POLLACTIVE] = 2,
634 [INFINIPATH_IBCS_LT_STATE_POLLQUIET] = 2,
635 [INFINIPATH_IBCS_LT_STATE_SLEEPDELAY] = 1,
636 [INFINIPATH_IBCS_LT_STATE_SLEEPQUIET] = 1,
637 [INFINIPATH_IBCS_LT_STATE_CFGDEBOUNCE] = 4,
638 [INFINIPATH_IBCS_LT_STATE_CFGRCVFCFG] = 4,
639 [INFINIPATH_IBCS_LT_STATE_CFGWAITRMT] = 4,
640 [INFINIPATH_IBCS_LT_STATE_CFGIDLE] = 4,
641 [INFINIPATH_IBCS_LT_STATE_RECOVERRETRAIN] = 6,
642 [INFINIPATH_IBCS_LT_STATE_RECOVERWAITRMT] = 6,
643 [INFINIPATH_IBCS_LT_STATE_RECOVERIDLE] = 6,
644};
645
646static int ipath_query_port(struct ib_device *ibdev,
647 u8 port, struct ib_port_attr *props)
648{
649 struct ipath_ibdev *dev = to_idev(ibdev);
650 enum ib_mtu mtu;
651 u16 lid = ipath_layer_get_lid(dev->dd);
652 u64 ibcstat;
653
654 memset(props, 0, sizeof(*props));
655 props->lid = lid ? lid : __constant_be16_to_cpu(IB_LID_PERMISSIVE);
656 props->lmc = dev->mkeyprot_resv_lmc & 7;
657 props->sm_lid = dev->sm_lid;
658 props->sm_sl = dev->sm_sl;
659 ibcstat = ipath_layer_get_lastibcstat(dev->dd);
660 props->state = ((ibcstat >> 4) & 0x3) + 1;
661 /* See phys_state_show() */
662 props->phys_state = ipath_cvt_physportstate[
663 ipath_layer_get_lastibcstat(dev->dd) & 0xf];
664 props->port_cap_flags = dev->port_cap_flags;
665 props->gid_tbl_len = 1;
c100f622 666 props->max_msg_sz = 0x80000000;
6522108f
BS
667 props->pkey_tbl_len = ipath_layer_get_npkeys(dev->dd);
668 props->bad_pkey_cntr = ipath_layer_get_cr_errpkey(dev->dd) -
443a64ab 669 dev->z_pkey_violations;
6522108f
BS
670 props->qkey_viol_cntr = dev->qkey_violations;
671 props->active_width = IB_WIDTH_4X;
672 /* See rate_show() */
673 props->active_speed = 1; /* Regular 10Mbs speed. */
674 props->max_vl_num = 1; /* VLCap = VL0 */
675 props->init_type_reply = 0;
676
677 props->max_mtu = IB_MTU_4096;
678 switch (ipath_layer_get_ibmtu(dev->dd)) {
679 case 4096:
680 mtu = IB_MTU_4096;
681 break;
682 case 2048:
683 mtu = IB_MTU_2048;
684 break;
685 case 1024:
686 mtu = IB_MTU_1024;
687 break;
688 case 512:
689 mtu = IB_MTU_512;
690 break;
691 case 256:
692 mtu = IB_MTU_256;
693 break;
694 default:
695 mtu = IB_MTU_2048;
696 }
697 props->active_mtu = mtu;
698 props->subnet_timeout = dev->subnet_timeout;
699
700 return 0;
701}
702
703static int ipath_modify_device(struct ib_device *device,
704 int device_modify_mask,
705 struct ib_device_modify *device_modify)
706{
707 int ret;
708
709 if (device_modify_mask & ~(IB_DEVICE_MODIFY_SYS_IMAGE_GUID |
710 IB_DEVICE_MODIFY_NODE_DESC)) {
711 ret = -EOPNOTSUPP;
712 goto bail;
713 }
714
715 if (device_modify_mask & IB_DEVICE_MODIFY_NODE_DESC)
716 memcpy(device->node_desc, device_modify->node_desc, 64);
717
718 if (device_modify_mask & IB_DEVICE_MODIFY_SYS_IMAGE_GUID)
719 to_idev(device)->sys_image_guid =
720 cpu_to_be64(device_modify->sys_image_guid);
721
722 ret = 0;
723
724bail:
725 return ret;
726}
727
728static int ipath_modify_port(struct ib_device *ibdev,
729 u8 port, int port_modify_mask,
730 struct ib_port_modify *props)
731{
732 struct ipath_ibdev *dev = to_idev(ibdev);
733
734 dev->port_cap_flags |= props->set_port_cap_mask;
735 dev->port_cap_flags &= ~props->clr_port_cap_mask;
736 if (port_modify_mask & IB_PORT_SHUTDOWN)
737 ipath_layer_set_linkstate(dev->dd, IPATH_IB_LINKDOWN);
738 if (port_modify_mask & IB_PORT_RESET_QKEY_CNTR)
739 dev->qkey_violations = 0;
740 return 0;
741}
742
743static int ipath_query_gid(struct ib_device *ibdev, u8 port,
744 int index, union ib_gid *gid)
745{
746 struct ipath_ibdev *dev = to_idev(ibdev);
747 int ret;
748
749 if (index >= 1) {
750 ret = -EINVAL;
751 goto bail;
752 }
753 gid->global.subnet_prefix = dev->gid_prefix;
754 gid->global.interface_id = ipath_layer_get_guid(dev->dd);
755
756 ret = 0;
757
758bail:
759 return ret;
760}
761
762static struct ib_pd *ipath_alloc_pd(struct ib_device *ibdev,
763 struct ib_ucontext *context,
764 struct ib_udata *udata)
765{
fe62546a 766 struct ipath_ibdev *dev = to_idev(ibdev);
6522108f
BS
767 struct ipath_pd *pd;
768 struct ib_pd *ret;
769
fe62546a
BS
770 /*
771 * This is actually totally arbitrary. Some correctness tests
772 * assume there's a maximum number of PDs that can be allocated.
773 * We don't actually have this limit, but we fail the test if
774 * we allow allocations of more than we report for this value.
775 */
776
c27fef26
BS
777 pd = kmalloc(sizeof *pd, GFP_KERNEL);
778 if (!pd) {
fe62546a
BS
779 ret = ERR_PTR(-ENOMEM);
780 goto bail;
781 }
782
c27fef26
BS
783 spin_lock(&dev->n_pds_lock);
784 if (dev->n_pds_allocated == ib_ipath_max_pds) {
785 spin_unlock(&dev->n_pds_lock);
786 kfree(pd);
6522108f
BS
787 ret = ERR_PTR(-ENOMEM);
788 goto bail;
789 }
790
fe62546a 791 dev->n_pds_allocated++;
c27fef26 792 spin_unlock(&dev->n_pds_lock);
fe62546a 793
6522108f
BS
794 /* ib_alloc_pd() will initialize pd->ibpd. */
795 pd->user = udata != NULL;
796
797 ret = &pd->ibpd;
798
799bail:
800 return ret;
801}
802
803static int ipath_dealloc_pd(struct ib_pd *ibpd)
804{
805 struct ipath_pd *pd = to_ipd(ibpd);
fe62546a
BS
806 struct ipath_ibdev *dev = to_idev(ibpd->device);
807
c27fef26 808 spin_lock(&dev->n_pds_lock);
fe62546a 809 dev->n_pds_allocated--;
c27fef26 810 spin_unlock(&dev->n_pds_lock);
6522108f
BS
811
812 kfree(pd);
813
814 return 0;
815}
816
817/**
818 * ipath_create_ah - create an address handle
819 * @pd: the protection domain
820 * @ah_attr: the attributes of the AH
821 *
822 * This may be called from interrupt context.
823 */
824static struct ib_ah *ipath_create_ah(struct ib_pd *pd,
825 struct ib_ah_attr *ah_attr)
826{
827 struct ipath_ah *ah;
828 struct ib_ah *ret;
fe62546a
BS
829 struct ipath_ibdev *dev = to_idev(pd->device);
830
6522108f 831 /* A multicast address requires a GRH (see ch. 8.4.1). */
27b678dd
BS
832 if (ah_attr->dlid >= IPATH_MULTICAST_LID_BASE &&
833 ah_attr->dlid != IPATH_PERMISSIVE_LID &&
6522108f
BS
834 !(ah_attr->ah_flags & IB_AH_GRH)) {
835 ret = ERR_PTR(-EINVAL);
836 goto bail;
837 }
838
4a45b7d4
BS
839 if (ah_attr->dlid == 0) {
840 ret = ERR_PTR(-EINVAL);
841 goto bail;
842 }
843
fe62546a 844 if (ah_attr->port_num < 1 ||
4a45b7d4
BS
845 ah_attr->port_num > pd->device->phys_port_cnt) {
846 ret = ERR_PTR(-EINVAL);
847 goto bail;
848 }
849
6522108f
BS
850 ah = kmalloc(sizeof *ah, GFP_ATOMIC);
851 if (!ah) {
852 ret = ERR_PTR(-ENOMEM);
853 goto bail;
854 }
855
c27fef26
BS
856 spin_lock(&dev->n_ahs_lock);
857 if (dev->n_ahs_allocated == ib_ipath_max_ahs) {
858 spin_unlock(&dev->n_ahs_lock);
859 kfree(ah);
860 ret = ERR_PTR(-ENOMEM);
861 goto bail;
862 }
863
fe62546a 864 dev->n_ahs_allocated++;
c27fef26 865 spin_unlock(&dev->n_ahs_lock);
fe62546a 866
6522108f
BS
867 /* ib_create_ah() will initialize ah->ibah. */
868 ah->attr = *ah_attr;
869
870 ret = &ah->ibah;
871
872bail:
873 return ret;
874}
875
876/**
877 * ipath_destroy_ah - destroy an address handle
878 * @ibah: the AH to destroy
879 *
880 * This may be called from interrupt context.
881 */
882static int ipath_destroy_ah(struct ib_ah *ibah)
883{
fe62546a 884 struct ipath_ibdev *dev = to_idev(ibah->device);
6522108f
BS
885 struct ipath_ah *ah = to_iah(ibah);
886
c27fef26 887 spin_lock(&dev->n_ahs_lock);
fe62546a 888 dev->n_ahs_allocated--;
c27fef26 889 spin_unlock(&dev->n_ahs_lock);
fe62546a 890
6522108f
BS
891 kfree(ah);
892
893 return 0;
894}
895
896static int ipath_query_ah(struct ib_ah *ibah, struct ib_ah_attr *ah_attr)
897{
898 struct ipath_ah *ah = to_iah(ibah);
899
900 *ah_attr = ah->attr;
901
902 return 0;
903}
904
905static int ipath_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
906 u16 *pkey)
907{
908 struct ipath_ibdev *dev = to_idev(ibdev);
909 int ret;
910
911 if (index >= ipath_layer_get_npkeys(dev->dd)) {
912 ret = -EINVAL;
913 goto bail;
914 }
915
916 *pkey = ipath_layer_get_pkey(dev->dd, index);
917 ret = 0;
918
919bail:
920 return ret;
921}
922
923
924/**
925 * ipath_alloc_ucontext - allocate a ucontest
926 * @ibdev: the infiniband device
927 * @udata: not used by the InfiniPath driver
928 */
929
930static struct ib_ucontext *ipath_alloc_ucontext(struct ib_device *ibdev,
931 struct ib_udata *udata)
932{
933 struct ipath_ucontext *context;
934 struct ib_ucontext *ret;
935
936 context = kmalloc(sizeof *context, GFP_KERNEL);
937 if (!context) {
938 ret = ERR_PTR(-ENOMEM);
939 goto bail;
940 }
941
942 ret = &context->ibucontext;
943
944bail:
945 return ret;
946}
947
948static int ipath_dealloc_ucontext(struct ib_ucontext *context)
949{
950 kfree(to_iucontext(context));
951 return 0;
952}
953
954static int ipath_verbs_register_sysfs(struct ib_device *dev);
955
956/**
957 * ipath_register_ib_device - register our device with the infiniband core
6522108f
BS
958 * @dd: the device data structure
959 * Return the allocated ipath_ibdev pointer or NULL on error.
960 */
b1c1b6a3 961int ipath_register_ib_device(struct ipath_devdata *dd)
6522108f 962{
fba75200 963 struct ipath_layer_counters cntrs;
6522108f
BS
964 struct ipath_ibdev *idev;
965 struct ib_device *dev;
966 int ret;
967
968 idev = (struct ipath_ibdev *)ib_alloc_device(sizeof *idev);
b1c1b6a3
BS
969 if (idev == NULL) {
970 ret = -ENOMEM;
6522108f 971 goto bail;
b1c1b6a3 972 }
6522108f
BS
973
974 dev = &idev->ibdev;
975
976 /* Only need to initialize non-zero fields. */
c27fef26
BS
977 spin_lock_init(&idev->n_pds_lock);
978 spin_lock_init(&idev->n_ahs_lock);
979 spin_lock_init(&idev->n_cqs_lock);
980 spin_lock_init(&idev->n_srqs_lock);
981 spin_lock_init(&idev->n_mcast_grps_lock);
982
6522108f
BS
983 spin_lock_init(&idev->qp_table.lock);
984 spin_lock_init(&idev->lk_table.lock);
985 idev->sm_lid = __constant_be16_to_cpu(IB_LID_PERMISSIVE);
986 /* Set the prefix to the default value (see ch. 4.1.1) */
987 idev->gid_prefix = __constant_cpu_to_be64(0xfe80000000000000ULL);
988
989 ret = ipath_init_qp_table(idev, ib_ipath_qp_table_size);
990 if (ret)
991 goto err_qp;
992
993 /*
994 * The top ib_ipath_lkey_table_size bits are used to index the
995 * table. The lower 8 bits can be owned by the user (copied from
996 * the LKEY). The remaining bits act as a generation number or tag.
997 */
998 idev->lk_table.max = 1 << ib_ipath_lkey_table_size;
999 idev->lk_table.table = kzalloc(idev->lk_table.max *
1000 sizeof(*idev->lk_table.table),
1001 GFP_KERNEL);
1002 if (idev->lk_table.table == NULL) {
1003 ret = -ENOMEM;
1004 goto err_lk;
1005 }
1006 spin_lock_init(&idev->pending_lock);
1007 INIT_LIST_HEAD(&idev->pending[0]);
1008 INIT_LIST_HEAD(&idev->pending[1]);
1009 INIT_LIST_HEAD(&idev->pending[2]);
1010 INIT_LIST_HEAD(&idev->piowait);
1011 INIT_LIST_HEAD(&idev->rnrwait);
1012 idev->pending_index = 0;
1013 idev->port_cap_flags =
1014 IB_PORT_SYS_IMAGE_GUID_SUP | IB_PORT_CLIENT_REG_SUP;
1015 idev->pma_counter_select[0] = IB_PMA_PORT_XMIT_DATA;
1016 idev->pma_counter_select[1] = IB_PMA_PORT_RCV_DATA;
1017 idev->pma_counter_select[2] = IB_PMA_PORT_XMIT_PKTS;
1018 idev->pma_counter_select[3] = IB_PMA_PORT_RCV_PKTS;
1019 idev->pma_counter_select[5] = IB_PMA_PORT_XMIT_WAIT;
1020 idev->link_width_enabled = 3; /* 1x or 4x */
1021
fba75200
BS
1022 /* Snapshot current HW counters to "clear" them. */
1023 ipath_layer_get_counters(dd, &cntrs);
1024 idev->z_symbol_error_counter = cntrs.symbol_error_counter;
1025 idev->z_link_error_recovery_counter =
1026 cntrs.link_error_recovery_counter;
1027 idev->z_link_downed_counter = cntrs.link_downed_counter;
1028 idev->z_port_rcv_errors = cntrs.port_rcv_errors;
1029 idev->z_port_rcv_remphys_errors =
1030 cntrs.port_rcv_remphys_errors;
1031 idev->z_port_xmit_discards = cntrs.port_xmit_discards;
1032 idev->z_port_xmit_data = cntrs.port_xmit_data;
1033 idev->z_port_rcv_data = cntrs.port_rcv_data;
1034 idev->z_port_xmit_packets = cntrs.port_xmit_packets;
1035 idev->z_port_rcv_packets = cntrs.port_rcv_packets;
1036 idev->z_local_link_integrity_errors =
1037 cntrs.local_link_integrity_errors;
1038 idev->z_excessive_buffer_overrun_errors =
1039 cntrs.excessive_buffer_overrun_errors;
1040
6522108f
BS
1041 /*
1042 * The system image GUID is supposed to be the same for all
1043 * IB HCAs in a single system but since there can be other
1044 * device types in the system, we can't be sure this is unique.
1045 */
1046 if (!sys_image_guid)
1047 sys_image_guid = ipath_layer_get_guid(dd);
1048 idev->sys_image_guid = sys_image_guid;
b1c1b6a3 1049 idev->ib_unit = dd->ipath_unit;
6522108f
BS
1050 idev->dd = dd;
1051
1052 strlcpy(dev->name, "ipath%d", IB_DEVICE_NAME_MAX);
41c75a19 1053 dev->owner = THIS_MODULE;
6522108f
BS
1054 dev->node_guid = ipath_layer_get_guid(dd);
1055 dev->uverbs_abi_ver = IPATH_UVERBS_ABI_VERSION;
1056 dev->uverbs_cmd_mask =
1057 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
1058 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
1059 (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
1060 (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
1061 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
1062 (1ull << IB_USER_VERBS_CMD_CREATE_AH) |
1063 (1ull << IB_USER_VERBS_CMD_DESTROY_AH) |
1064 (1ull << IB_USER_VERBS_CMD_QUERY_AH) |
1065 (1ull << IB_USER_VERBS_CMD_REG_MR) |
1066 (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
1067 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
1068 (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
1069 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ) |
1070 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
1071 (1ull << IB_USER_VERBS_CMD_POLL_CQ) |
1072 (1ull << IB_USER_VERBS_CMD_REQ_NOTIFY_CQ) |
1073 (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
1074 (1ull << IB_USER_VERBS_CMD_QUERY_QP) |
1075 (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
1076 (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
1077 (1ull << IB_USER_VERBS_CMD_POST_SEND) |
1078 (1ull << IB_USER_VERBS_CMD_POST_RECV) |
1079 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST) |
1080 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST) |
1081 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ) |
1082 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) |
1083 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ) |
1084 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ) |
1085 (1ull << IB_USER_VERBS_CMD_POST_SRQ_RECV);
1086 dev->node_type = IB_NODE_CA;
1087 dev->phys_port_cnt = 1;
1088 dev->dma_device = ipath_layer_get_device(dd);
1089 dev->class_dev.dev = dev->dma_device;
1090 dev->query_device = ipath_query_device;
1091 dev->modify_device = ipath_modify_device;
1092 dev->query_port = ipath_query_port;
1093 dev->modify_port = ipath_modify_port;
1094 dev->query_pkey = ipath_query_pkey;
1095 dev->query_gid = ipath_query_gid;
1096 dev->alloc_ucontext = ipath_alloc_ucontext;
1097 dev->dealloc_ucontext = ipath_dealloc_ucontext;
1098 dev->alloc_pd = ipath_alloc_pd;
1099 dev->dealloc_pd = ipath_dealloc_pd;
1100 dev->create_ah = ipath_create_ah;
1101 dev->destroy_ah = ipath_destroy_ah;
1102 dev->query_ah = ipath_query_ah;
1103 dev->create_srq = ipath_create_srq;
1104 dev->modify_srq = ipath_modify_srq;
1105 dev->query_srq = ipath_query_srq;
1106 dev->destroy_srq = ipath_destroy_srq;
1107 dev->create_qp = ipath_create_qp;
1108 dev->modify_qp = ipath_modify_qp;
1109 dev->query_qp = ipath_query_qp;
1110 dev->destroy_qp = ipath_destroy_qp;
1111 dev->post_send = ipath_post_send;
1112 dev->post_recv = ipath_post_receive;
1113 dev->post_srq_recv = ipath_post_srq_receive;
1114 dev->create_cq = ipath_create_cq;
1115 dev->destroy_cq = ipath_destroy_cq;
1116 dev->resize_cq = ipath_resize_cq;
1117 dev->poll_cq = ipath_poll_cq;
1118 dev->req_notify_cq = ipath_req_notify_cq;
1119 dev->get_dma_mr = ipath_get_dma_mr;
1120 dev->reg_phys_mr = ipath_reg_phys_mr;
1121 dev->reg_user_mr = ipath_reg_user_mr;
1122 dev->dereg_mr = ipath_dereg_mr;
1123 dev->alloc_fmr = ipath_alloc_fmr;
1124 dev->map_phys_fmr = ipath_map_phys_fmr;
1125 dev->unmap_fmr = ipath_unmap_fmr;
1126 dev->dealloc_fmr = ipath_dealloc_fmr;
1127 dev->attach_mcast = ipath_multicast_attach;
1128 dev->detach_mcast = ipath_multicast_detach;
1129 dev->process_mad = ipath_process_mad;
373d9915 1130 dev->mmap = ipath_mmap;
6522108f
BS
1131
1132 snprintf(dev->node_desc, sizeof(dev->node_desc),
1133 IPATH_IDSTR " %s kernel_SMA", system_utsname.nodename);
1134
1135 ret = ib_register_device(dev);
1136 if (ret)
1137 goto err_reg;
1138
1139 if (ipath_verbs_register_sysfs(dev))
1140 goto err_class;
1141
1142 ipath_layer_enable_timer(dd);
1143
1144 goto bail;
1145
1146err_class:
1147 ib_unregister_device(dev);
1148err_reg:
1149 kfree(idev->lk_table.table);
1150err_lk:
1151 kfree(idev->qp_table.table);
1152err_qp:
1153 ib_dealloc_device(dev);
1154 _VERBS_ERROR("ib_ipath%d cannot register verbs (%d)!\n",
b1c1b6a3 1155 dd->ipath_unit, -ret);
6522108f
BS
1156 idev = NULL;
1157
1158bail:
b1c1b6a3
BS
1159 dd->verbs_dev = idev;
1160 return ret;
6522108f
BS
1161}
1162
b1c1b6a3 1163void ipath_unregister_ib_device(struct ipath_ibdev *dev)
6522108f 1164{
6522108f
BS
1165 struct ib_device *ibdev = &dev->ibdev;
1166
1167 ipath_layer_disable_timer(dev->dd);
1168
1169 ib_unregister_device(ibdev);
1170
1171 if (!list_empty(&dev->pending[0]) ||
1172 !list_empty(&dev->pending[1]) ||
1173 !list_empty(&dev->pending[2]))
1174 _VERBS_ERROR("ipath%d pending list not empty!\n",
1175 dev->ib_unit);
1176 if (!list_empty(&dev->piowait))
1177 _VERBS_ERROR("ipath%d piowait list not empty!\n",
1178 dev->ib_unit);
1179 if (!list_empty(&dev->rnrwait))
1180 _VERBS_ERROR("ipath%d rnrwait list not empty!\n",
1181 dev->ib_unit);
1182 if (!ipath_mcast_tree_empty())
1183 _VERBS_ERROR("ipath%d multicast table memory leak!\n",
1184 dev->ib_unit);
1185 /*
1186 * Note that ipath_unregister_ib_device() can be called before all
1187 * the QPs are destroyed!
1188 */
1189 ipath_free_all_qps(&dev->qp_table);
1190 kfree(dev->qp_table.table);
1191 kfree(dev->lk_table.table);
1192 ib_dealloc_device(ibdev);
1193}
1194
6522108f
BS
1195static ssize_t show_rev(struct class_device *cdev, char *buf)
1196{
5494c22b
RD
1197 struct ipath_ibdev *dev =
1198 container_of(cdev, struct ipath_ibdev, ibdev.class_dev);
6522108f 1199
e8a88f09 1200 return sprintf(buf, "%x\n", ipath_layer_get_pcirev(dev->dd));
6522108f
BS
1201}
1202
1203static ssize_t show_hca(struct class_device *cdev, char *buf)
1204{
5494c22b
RD
1205 struct ipath_ibdev *dev =
1206 container_of(cdev, struct ipath_ibdev, ibdev.class_dev);
1207 int ret;
6522108f 1208
5494c22b
RD
1209 ret = ipath_layer_get_boardname(dev->dd, buf, 128);
1210 if (ret < 0)
1211 goto bail;
1212 strcat(buf, "\n");
1213 ret = strlen(buf);
6522108f
BS
1214
1215bail:
1216 return ret;
1217}
1218
1219static ssize_t show_stats(struct class_device *cdev, char *buf)
1220{
5494c22b
RD
1221 struct ipath_ibdev *dev =
1222 container_of(cdev, struct ipath_ibdev, ibdev.class_dev);
1223 int i;
1224 int len;
1225
1226 len = sprintf(buf,
1227 "RC resends %d\n"
9b2017f1 1228 "RC no QACK %d\n"
5494c22b
RD
1229 "RC ACKs %d\n"
1230 "RC SEQ NAKs %d\n"
1231 "RC RDMA seq %d\n"
1232 "RC RNR NAKs %d\n"
1233 "RC OTH NAKs %d\n"
1234 "RC timeouts %d\n"
1235 "RC RDMA dup %d\n"
1236 "piobuf wait %d\n"
1237 "no piobuf %d\n"
1238 "PKT drops %d\n"
1239 "WQE errs %d\n",
1240 dev->n_rc_resends, dev->n_rc_qacks, dev->n_rc_acks,
1241 dev->n_seq_naks, dev->n_rdma_seq, dev->n_rnr_naks,
1242 dev->n_other_naks, dev->n_timeouts,
1243 dev->n_rdma_dup_busy, dev->n_piowait,
1244 dev->n_no_piobuf, dev->n_pkt_drops, dev->n_wqe_errs);
1245 for (i = 0; i < ARRAY_SIZE(dev->opstats); i++) {
6522108f
BS
1246 const struct ipath_opcode_stats *si = &dev->opstats[i];
1247
5494c22b
RD
1248 if (!si->n_packets && !si->n_bytes)
1249 continue;
1250 len += sprintf(buf + len, "%02x %llu/%llu\n", i,
6522108f 1251 (unsigned long long) si->n_packets,
5494c22b
RD
1252 (unsigned long long) si->n_bytes);
1253 }
1254 return len;
6522108f
BS
1255}
1256
1257static CLASS_DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
1258static CLASS_DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
1259static CLASS_DEVICE_ATTR(board_id, S_IRUGO, show_hca, NULL);
1260static CLASS_DEVICE_ATTR(stats, S_IRUGO, show_stats, NULL);
1261
1262static struct class_device_attribute *ipath_class_attributes[] = {
5494c22b
RD
1263 &class_device_attr_hw_rev,
1264 &class_device_attr_hca_type,
1265 &class_device_attr_board_id,
1266 &class_device_attr_stats
6522108f
BS
1267};
1268
1269static int ipath_verbs_register_sysfs(struct ib_device *dev)
1270{
5494c22b 1271 int i;
6522108f
BS
1272 int ret;
1273
5494c22b
RD
1274 for (i = 0; i < ARRAY_SIZE(ipath_class_attributes); ++i)
1275 if (class_device_create_file(&dev->class_dev,
1276 ipath_class_attributes[i])) {
1277 ret = 1;
6522108f
BS
1278 goto bail;
1279 }
1280
5494c22b 1281 ret = 0;
6522108f
BS
1282
1283bail:
1284 return ret;
1285}