aboutsummaryrefslogtreecommitdiff
path: root/drivers/infiniband/hw/ipath/ipath_ruc.c
blob: 54c61a972de24153ad7940ba93f080d190765646 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
/*
 * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
 * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
 *
 * This software is available to you under a choice of one of two
 * licenses.  You may choose to be licensed under the terms of the GNU
 * General Public License (GPL) Version 2, available from the file
 * COPYING in the main directory of this source tree, or the
 * OpenIB.org BSD license below:
 *
 *     Redistribution and use in source and binary forms, with or
 *     without modification, are permitted provided that the following
 *     conditions are met:
 *
 *      - Redistributions of source code must retain the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer.
 *
 *      - Redistributions in binary form must reproduce the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer in the documentation and/or other materials
 *        provided with the distribution.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 */

#include <linux/spinlock.h>

#include "ipath_verbs.h"
#include "ipath_kernel.h"

/*
 * Convert the AETH RNR timeout code into the number of milliseconds.
 */
const u32 ib_ipath_rnr_table[32] = {
	656,			/* 0 */
	1,			/* 1 */
	1,			/* 2 */
	1,			/* 3 */
	1,			/* 4 */
	1,			/* 5 */
	1,			/* 6 */
	1,			/* 7 */
	1,			/* 8 */
	1,			/* 9 */
	1,			/* A */
	1,			/* B */
	1,			/* C */
	1,			/* D */
	2,			/* E */
	2,			/* F */
	3,			/* 10 */
	4,			/* 11 */
	6,			/* 12 */
	8,			/* 13 */
	11,			/* 14 */
	16,			/* 15 */
	21,			/* 16 */
	31,			/* 17 */
	41,			/* 18 */
	62,			/* 19 */
	82,			/* 1A */
	123,			/* 1B */
	164,			/* 1C */
	246,			/* 1D */
	328,			/* 1E */
	492			/* 1F */
};

/**
 * ipath_insert_rnr_queue - put QP on the RNR timeout list for the device
 * @qp: the QP
 *
 * XXX Use a simple list for now.  We might need a priority
 * queue if we have lots of QPs waiting for RNR timeouts
 * but that should be rare.
 */
void ipath_insert_rnr_queue(struct ipath_qp *qp)
{
	struct ipath_ibdev *dev = to_idev(qp->ibqp.device);
	unsigned long flags;

	spin_lock_irqsave(&dev->pending_lock, flags);
	if (list_empty(&dev->rnrwait))
		list_add(&qp->timerwait, &dev->rnrwait);
	else {
		struct list_head *l = &dev->rnrwait;
		struct ipath_qp *nqp = list_entry(l->next, struct ipath_qp,
						  timerwait);

		while (qp->s_rnr_timeout >= nqp->s_rnr_timeout) {
			qp->s_rnr_timeout -= nqp->s_rnr_timeout;
			l = l->next;
			if (l->next == &dev->rnrwait)
				break;
			nqp = list_entry(l->next, struct ipath_qp,
					 timerwait);
		}
		list_add(&qp->timerwait, l);
	}
	spin_unlock_irqrestore(&dev->pending_lock, flags);
}

/**
 * ipath_init_sge - Validate a RWQE and fill in the SGE state
 * @qp: the QP
 *
 * Return 1 if OK.
 */
int ipath_init_sge(struct ipath_qp *qp, struct ipath_rwqe *wqe,
		   u32 *lengthp, struct ipath_sge_state *ss)
{
	int i, j, ret;
	struct ib_wc wc;

	*lengthp = 0;
	for (i = j = 0; i < wqe->num_sge; i++) {
		if (wqe->sg_list[i].length == 0)
			continue;
		/* Check LKEY */
		if (!ipath_lkey_ok(qp, j ? &ss->sg_list[j - 1] : &ss->sge,
				   &wqe->sg_list[i], IB_ACCESS_LOCAL_WRITE))
			goto bad_lkey;
		*lengthp += wqe->sg_list[i].length;
		j++;
	}
	ss->num_sge = j;
	ret = 1;
	goto bail;

bad_lkey:
	wc.wr_id = wqe->wr_id;
	wc.status = IB_WC_LOC_PROT_ERR;
	wc.opcode = IB_WC_RECV;
	wc.vendor_err = 0;
	wc.byte_len = 0;
	wc.imm_data = 0;
	wc.qp = &qp->ibqp;
	wc.src_qp = 0;
	wc.wc_flags = 0;
	wc.pkey_index = 0;
	wc.slid = 0;
	wc.sl = 0;
	wc.dlid_path_bits = 0;
	wc.port_num = 0;
	/* Signal solicited completion event. */
	ipath_cq_enter(to_icq(qp->ibqp.recv_cq), &wc, 1);
	ret = 0;
bail:
	return ret;
}

/**
 * ipath_get_rwqe - copy the next RWQE into the QP's RWQE
 * @qp: the QP
 * @wr_id_only: update wr_id only, not SGEs
 *
 * Return 0 if no RWQE is available, otherwise return 1.
 *
 * Can be called from interrupt level.
 */
int ipath_get_rwqe(struct ipath_qp *qp, int wr_id_only)
{
	unsigned long flags;
	struct ipath_rq *rq;
	struct ipath_rwq *wq;
	struct ipath_srq *srq;
	struct ipath_rwqe *wqe;
	void (*handler)(struct ib_event *, void *);
	u32 tail;
	int ret;

	qp->r_sge.sg_list = qp->r_sg_list;

	if (qp->ibqp.srq) {
		srq = to_isrq(qp->ibqp.srq);
		handler = srq->ibsrq.event_handler;
		rq = &srq->rq;
	} else {
		srq = NULL;
		handler = NULL;
		rq = &qp->r_rq;
	}

	spin_lock_irqsave(&rq->lock, flags);
	wq = rq->wq;
	tail = wq->tail;
	/* Validate tail before using it since it is user writable. */
	if (tail >= rq->size)
		tail = 0;
	do {
		if (unlikely(tail == wq->head)) {
			spin_unlock_irqrestore(&rq->lock, flags);
			ret = 0;
			goto bail;
		}
		/* Make sure entry is read after head index is read. */
		smp_rmb();
		wqe = get_rwqe_ptr(rq, tail);
		if (++tail >= rq->size)
			tail = 0;
	} while (!wr_id_only && !ipath_init_sge(qp, wqe, &qp->r_len,
						&qp->r_sge));
	qp->r_wr_id = wqe->wr_id;
	wq->tail = tail;

	ret = 1;
	qp->r_wrid_valid = 1;
	if (handler) {
		u32 n;

		/*
		 * validate head pointer value and compute
		 * the number of remaining WQEs.
		 */
		n = wq->head;
		if (n >= rq->size)
			n = 0;
		if (n < tail)
			n += rq->size - tail;
		else
			n -= tail;
		if (n < srq->limit) {
			struct ib_event ev;

			srq->limit = 0;
			spin_unlock_irqrestore(&rq->lock, flags);
			ev.device = qp->ibqp.device;
			ev.element.srq = qp->ibqp.srq;
			ev.event = IB_EVENT_SRQ_LIMIT_REACHED;
			handler(&ev, srq->ibsrq.srq_context);
			goto bail;
		}
	}
	spin_unlock_irqrestore(&rq->lock, flags);

bail:
	return ret;
}

/**
 * ipath_ruc_loopback - handle UC and RC lookback requests
 * @sqp: the sending QP
 *
 * This is called from ipath_do_send() to
 * forward a WQE addressed to the same HCA.
 * Note that although we are single threaded due to the tasklet, we still
 * have to protect against post_send().  We don't have to worry about
 * receive interrupts since this is a connected protocol and all packets
 * will pass through here.
 */
static void ipath_ruc_loopback(struct ipath_qp *sqp)
{
	struct ipath_ibdev *dev = to_idev(sqp->ibqp.device);
	struct ipath_qp *qp;
	struct ipath_swqe *wqe;
	struct ipath_sge *sge;
	unsigned long flags;
	struct ib_wc wc;
	u64 sdata;
	atomic64_t *maddr;

	qp = ipath_lookup_qpn(&dev->qp_table, sqp->remote_qpn);
	if (!qp) {
		dev->n_pkt_drops++;
		return;
	}

again:
	spin_lock_irqsave(&sqp->s_lock, flags);

	if (!(ib_ipath_state_ops[sqp->state] & IPATH_PROCESS_SEND_OK) ||
	    sqp->s_rnr_timeout) {
		spin_unlock_irqrestore(&sqp->s_lock, flags);
		goto done;
	}

	/* Get the next send request. */
	if (sqp->s_last == sqp->s_head) {
		/* Send work queue is empty. */
		spin_unlock_irqrestore(&sqp->s_lock, flags);
		goto done;
	}

	/*
	 * We can rely on the entry not changing without the s_lock
	 * being held until we update s_last.
	 */
	wqe = get_swqe_ptr(sqp, sqp->s_last);
	spin_unlock_irqrestore(&sqp->s_lock, flags);

	wc.wc_flags = 0;
	wc.imm_data = 0;

	sqp->s_sge.sge = wqe->sg_list[0];
	sqp->s_sge.sg_list = wqe->sg_list + 1;
	sqp->s_sge.num_sge = wqe->wr.num_sge;
	sqp->s_len = wqe->length;
	switch (wqe->wr.opcode) {
	case IB_WR_SEND_WITH_IMM:
		wc.wc_flags = IB_WC_WITH_IMM;
		wc.imm_data = wqe->wr.imm_data;
		/* FALLTHROUGH */
	case IB_WR_SEND:
		if (!ipath_get_rwqe(qp, 0)) {
		rnr_nak:
			/* Handle RNR NAK */
			if (qp->ibqp.qp_type == IB_QPT_UC)
				goto send_comp;
			if (sqp->s_rnr_retry == 0) {
				wc.status = IB_WC_RNR_RETRY_EXC_ERR;
				goto err;
			}
			if (sqp->s_rnr_retry_cnt < 7)
				sqp->s_rnr_retry--;
			dev->n_rnr_naks++;
			sqp->s_rnr_timeout =
				ib_ipath_rnr_table[qp->r_min_rnr_timer];
			ipath_insert_rnr_queue(sqp);
			goto done;
		}
		break;

	case IB_WR_RDMA_WRITE_WITH_IMM:
		if (unlikely(!(qp->qp_access_flags &
			       IB_ACCESS_REMOTE_WRITE))) {
			wc.status = IB_WC_REM_INV_REQ_ERR;
			goto err;
		}
		wc.wc_flags = IB_WC_WITH_IMM;
		wc.imm_data = wqe->wr.imm_data;
		if (!ipath_get_rwqe(qp, 1))
			goto rnr_nak;
		/* FALLTHROUGH */
	case IB_WR_RDMA_WRITE:
		if (unlikely(!(qp->qp_access_flags &
			       IB_ACCESS_REMOTE_WRITE))) {
			wc.status = IB_WC_REM_INV_REQ_ERR;
			goto err;
		}
		if (wqe->length == 0)
			break;
		if (unlikely(!ipath_rkey_ok(qp, &qp->r_sge, wqe->length,
					    wqe->wr.wr.rdma.remote_addr,
					    wqe->wr.wr.rdma.rkey,
					    IB_ACCESS_REMOTE_WRITE))) {
		acc_err:
			wc.status = IB_WC_REM_ACCESS_ERR;
		err:
			wc.wr_id = wqe->wr.wr_id;
			wc.opcode = ib_ipath_wc_opcode[wqe->wr.opcode];
			wc.vendor_err = 0;
			wc.byte_len = 0;
			wc.qp = &sqp->ibqp;
			wc.src_qp = sqp->remote_qpn;
			wc.pkey_index = 0;
			wc.slid = sqp->remote_ah_attr.dlid;
			wc.sl = sqp->remote_ah_attr.sl;
			wc.dlid_path_bits = 0;
			wc.port_num = 0;
			spin_lock_irqsave(&sqp->s_lock, flags);
			ipath_sqerror_qp(sqp, &wc);
			spin_unlock_irqrestore(&sqp->s_lock, flags);
			goto done;
		}
		break;

	case IB_WR_RDMA_READ:
		if (unlikely(!(qp->qp_access_flags &
			       IB_ACCESS_REMOTE_READ))) {
			wc.status = IB_WC_REM_INV_REQ_ERR;
			goto err;
		}
		if (unlikely(!ipath_rkey_ok(qp, &sqp->s_sge, wqe->length,
					    wqe->wr.wr.rdma.remote_addr,
					    wqe->wr.wr.rdma.rkey,
					    IB_ACCESS_REMOTE_READ)))
			goto acc_err;
		qp->r_sge.sge = wqe->sg_list[0];
		qp->r_sge.sg_list = wqe->sg_list + 1;
		qp->r_sge.num_sge = wqe->wr.num_sge;
		break;

	case IB_WR_ATOMIC_CMP_AND_SWP:
	case IB_WR_ATOMIC_FETCH_AND_ADD:
		if (unlikely(!(qp->qp_access_flags &
			       IB_ACCESS_REMOTE_ATOMIC))) {
			wc.status = IB_WC_REM_INV_REQ_ERR;
			goto err;
		}
		if (unlikely(!ipath_rkey_ok(qp, &qp->r_sge, sizeof(u64),
					    wqe->wr.wr.atomic.remote_addr,
					    wqe->wr.wr.atomic.rkey,
					    IB_ACCESS_REMOTE_ATOMIC)))
			goto acc_err;
		/* Perform atomic OP and save result. */
		maddr = (atomic64_t *) qp->r_sge.sge.vaddr;
		sdata = wqe->wr.wr.atomic.compare_add;
		*(u64 *) sqp->s_sge.sge.vaddr =
			(wqe->wr.opcode == IB_WR_ATOMIC_FETCH_AND_ADD) ?
			(u64) atomic64_add_return(sdata, maddr) - sdata :
			(u64) cmpxchg((u64 *) qp->r_sge.sge.vaddr,
				      sdata, wqe->wr.wr.atomic.swap);
		goto send_comp;

	default:
		goto done;
	}

	sge = &sqp->s_sge.sge;
	while (sqp->s_len) {
		u32 len = sqp->s_len;

		if (len > sge->length)
			len = sge->length;
		if (len > sge->sge_length)
			len = sge->sge_length;
		BUG_ON(len == 0);
		ipath_copy_sge(&qp->r_sge, sge->vaddr, len);
		sge->vaddr += len;
		sge->length -= len;
		sge->sge_length -= len;
		if (sge->sge_length == 0) {
			if (--sqp->s_sge.num_sge)
				*sge = *sqp->s_sge.sg_list++;
		} else if (sge->length == 0 && sge->mr != NULL) {
			if (++sge->n >= IPATH_SEGSZ) {
				if (++sge->m >= sge->mr->mapsz)
					break;
				sge->n = 0;
			}
			sge->vaddr =
				sge->mr->map[sge->m]->segs[sge->n].vaddr;
			sge->length =
				sge->mr->map[sge->m]->segs[sge->n].length;
		}
		sqp->s_len -= len;
	}

	if (wqe->wr.opcode == IB_WR_RDMA_WRITE ||
	    wqe->wr.opcode == IB_WR_RDMA_READ)
		goto send_comp;

	if (wqe->wr.opcode == IB_WR_RDMA_WRITE_WITH_IMM)
		wc.opcode = IB_WC_RECV_RDMA_WITH_IMM;
	else
		wc.opcode = IB_WC_RECV;
	wc.wr_id = qp->r_wr_id;
	wc.status = IB_WC_SUCCESS;
	wc.vendor_err = 0;
	wc.byte_len = wqe->length;
	wc.qp = &qp->ibqp;
	wc.src_qp = qp->remote_qpn;
	wc.pkey_index = 0;
	wc.slid = qp->remote_ah_attr.dlid;
	wc.sl = qp->remote_ah_attr.sl;
	wc.dlid_path_bits = 0;
	wc.port_num = 1;
	/* Signal completion event if the solicited bit is set. */
	ipath_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
		       wqe->wr.send_flags & IB_SEND_SOLICITED);

send_comp:
	sqp->s_rnr_retry = sqp->s_rnr_retry_cnt;
	ipath_send_complete(sqp, wqe, IB_WC_SUCCESS);
	goto again;

done:
	if (atomic_dec_and_test(&qp->refcount))
		wake_up(&qp->wait);
}

static void want_buffer(struct ipath_devdata *dd)
{
	set_bit(IPATH_S_PIOINTBUFAVAIL, &dd->ipath_sendctrl);
	ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl,
			 dd->ipath_sendctrl);
}

/**
 * ipath_no_bufs_available - tell the layer driver we need buffers
 * @qp: the QP that caused the problem
 * @dev: the device we ran out of buffers on
 *
 * Called when we run out of PIO buffers.
 */
static void ipath_no_bufs_available(struct ipath_qp *qp,
				    struct ipath_ibdev *dev)
{
	unsigned long flags;

	/*
	 * Note that as soon as want_buffer() is called and
	 * possibly before it returns, ipath_ib_piobufavail()
	 * could be called.  If we are still in the tasklet function,
	 * tasklet_hi_schedule() will not call us until the next time
	 * tasklet_hi_schedule() is called.
	 * We leave the busy flag set so that another post send doesn't
	 * try to put the same QP on the piowait list again.
	 */
	spin_lock_irqsave(&dev->pending_lock, flags);
	list_add_tail(&qp->piowait, &dev->piowait);
	spin_unlock_irqrestore(&dev->pending_lock, flags);
	want_buffer(dev->dd);
	dev->n_piowait++;
}

/**
 * ipath_make_grh - construct a GRH header
 * @dev: a pointer to the ipath device
 * @hdr: a pointer to the GRH header being constructed
 * @grh: the global route address to send to
 * @hwords: the number of 32 bit words of header being sent
 * @nwords: the number of 32 bit words of data being sent
 *
 * Return the size of the header in 32 bit words.
 */
u32 ipath_make_grh(struct ipath_ibdev *dev, struct ib_grh *hdr,
		   struct ib_global_route *grh, u32 hwords, u32 nwords)
{
	hdr->version_tclass_flow =
		cpu_to_be32((6 << 28) |
			    (grh->traffic_class << 20) |
			    grh->flow_label);
	hdr->paylen = cpu_to_be16((hwords - 2 + nwords + SIZE_OF_CRC) << 2);
	/* next_hdr is defined by C8-7 in ch. 8.4.1 */
	hdr->next_hdr = 0x1B;
	hdr->hop_limit = grh->hop_limit;
	/* The SGID is 32-bit aligned. */
	hdr->sgid.global.subnet_prefix = dev->gid_prefix;
	hdr->sgid.global.interface_id = dev->dd->ipath_guid;
	hdr->dgid = grh->dgid;

	/* GRH header size in 32-bit words. */
	return sizeof(struct ib_grh) / sizeof(u32);
}

void ipath_make_ruc_header(struct ipath_ibdev *dev, struct ipath_qp *qp,
			   struct ipath_other_headers *ohdr,
			   u32 bth0, u32 bth2)
{
	u16 lrh0;
	u32 nwords;
	u32 extra_bytes;

	/* Construct the header. */
	extra_bytes = -qp->s_cur_size & 3;
	nwords = (qp->s_cur_size + extra_bytes) >> 2;
	lrh0 = IPATH_LRH_BTH;
	if (unlikely(qp->remote_ah_attr.ah_flags & IB_AH_GRH)) {
		qp->s_hdrwords += ipath_make_grh(dev, &qp->s_hdr.u.l.grh,
						 &qp->remote_ah_attr.grh,
						 qp->s_hdrwords, nwords);
		lrh0 = IPATH_LRH_GRH;
	}
	lrh0 |= qp->remote_ah_attr.sl << 4;
	qp->s_hdr.lrh[0] = cpu_to_be16(lrh0);
	qp->s_hdr.lrh[1] = cpu_to_be16(qp->remote_ah_attr.dlid);
	qp->s_hdr.lrh[2] = cpu_to_be16(qp->s_hdrwords + nwords + SIZE_OF_CRC);
	qp->s_hdr.lrh[3] = cpu_to_be16(dev->dd->ipath_lid);
	bth0 |= ipath_get_pkey(dev->dd, qp->s_pkey_index);
	bth0 |= extra_bytes << 20;
	ohdr->bth[0] = cpu_to_be32(bth0 | (1 << 22));
	ohdr->bth[1] = cpu_to_be32(qp->remote_qpn);
	ohdr->bth[2] = cpu_to_be32(bth2);
}

/**
 * ipath_do_send - perform a send on a QP
 * @data: contains a pointer to the QP
 *
 * Process entries in the send work queue until credit or queue is
 * exhausted.  Only allow one CPU to send a packet per QP (tasklet).
 * Otherwise, two threads could send packets out of order.
 */
void ipath_do_send(unsigned long data)
{
	struct ipath_qp *qp = (struct ipath_qp *)data;
	struct ipath_ibdev *dev = to_idev(qp->ibqp.device);
	int (*make_req)(struct ipath_qp *qp);

	if (test_and_set_bit(IPATH_S_BUSY, &qp->s_busy))
		goto bail;

	if ((qp->ibqp.qp_type == IB_QPT_RC ||
	     qp->ibqp.qp_type == IB_QPT_UC) &&
	    qp->remote_ah_attr.dlid == dev->dd->ipath_lid) {
		ipath_ruc_loopback(qp);
		goto clear;
	}

	if (qp->ibqp.qp_type == IB_QPT_RC)
	       make_req = ipath_make_rc_req;
	else if (qp->ibqp.qp_type == IB_QPT_UC)
	       make_req = ipath_make_uc_req;
	else
	       make_req = ipath_make_ud_req;

again:
	/* Check for a constructed packet to be sent. */
	if (qp->s_hdrwords != 0) {
		/*
		 * If no PIO bufs are available, return.  An interrupt will
		 * call ipath_ib_piobufavail() when one is available.
		 */
		if (ipath_verbs_send(qp, &qp->s_hdr, qp->s_hdrwords,
				     qp->s_cur_sge, qp->s_cur_size)) {
			ipath_no_bufs_available(qp, dev);
			goto bail;
		}
		dev->n_unicast_xmit++;
		/* Record that we sent the packet and s_hdr is empty. */
		qp->s_hdrwords = 0;
	}

	if (make_req(qp))
		goto again;
clear:
	clear_bit(IPATH_S_BUSY, &qp->s_busy);
bail:;
}

void ipath_send_complete(struct ipath_qp *qp, struct ipath_swqe *wqe,
			 enum ib_wc_status status)
{
	unsigned long flags;
	u32 last;

	/* See ch. 11.2.4.1 and 10.7.3.1 */
	if (!(qp->s_flags & IPATH_S_SIGNAL_REQ_WR) ||
	    (wqe->wr.send_flags & IB_SEND_SIGNALED) ||
	    status != IB_WC_SUCCESS) {
		struct ib_wc wc;

		wc.wr_id = wqe->wr.wr_id;
		wc.status = status;
		wc.opcode = ib_ipath_wc_opcode[wqe->wr.opcode];
		wc.vendor_err = 0;
		wc.byte_len = wqe->length;
		wc.imm_data = 0;
		wc.qp = &qp->ibqp;
		wc.src_qp = 0;
		wc.wc_flags = 0;
		wc.pkey_index = 0;
		wc.slid = 0;
		wc.sl = 0;
		wc.dlid_path_bits = 0;
		wc.port_num = 0;
		ipath_cq_enter(to_icq(qp->ibqp.send_cq), &wc, 0);
	}

	spin_lock_irqsave(&qp->s_lock, flags);
	last = qp->s_last;
	if (++last >= qp->s_size)
		last = 0;
	qp->s_last = last;
	spin_unlock_irqrestore(&qp->s_lock, flags);
}