summaryrefslogtreecommitdiff
path: root/drivers/infiniband/hw/mthca/mthca_srq.c
blob: b8f05a52667376830b8dd450261341b49cb30fe8 (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
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
/*
 * Copyright (c) 2005 Cisco Systems. 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.
 *
 * $Id: mthca_srq.c 3047 2005-08-10 03:59:35Z roland $
 */

#include <linux/slab.h>
#include <linux/string.h>
#include <linux/sched.h>

#include <asm/io.h>

#include "mthca_dev.h"
#include "mthca_cmd.h"
#include "mthca_memfree.h"
#include "mthca_wqe.h"

enum {
	MTHCA_MAX_DIRECT_SRQ_SIZE = 4 * PAGE_SIZE
};

struct mthca_tavor_srq_context {
	__be64 wqe_base_ds;	/* low 6 bits is descriptor size */
	__be32 state_pd;
	__be32 lkey;
	__be32 uar;
	__be16 limit_watermark;
	__be16 wqe_cnt;
	u32    reserved[2];
};

struct mthca_arbel_srq_context {
	__be32 state_logsize_srqn;
	__be32 lkey;
	__be32 db_index;
	__be32 logstride_usrpage;
	__be64 wqe_base;
	__be32 eq_pd;
	__be16 limit_watermark;
	__be16 wqe_cnt;
	u16    reserved1;
	__be16 wqe_counter;
	u32    reserved2[3];
};

static void *get_wqe(struct mthca_srq *srq, int n)
{
	if (srq->is_direct)
		return srq->queue.direct.buf + (n << srq->wqe_shift);
	else
		return srq->queue.page_list[(n << srq->wqe_shift) >> PAGE_SHIFT].buf +
			((n << srq->wqe_shift) & (PAGE_SIZE - 1));
}

/*
 * Return a pointer to the location within a WQE that we're using as a
 * link when the WQE is in the free list.  We use the imm field
 * because in the Tavor case, posting a WQE may overwrite the next
 * segment of the previous WQE, but a receive WQE will never touch the
 * imm field.  This avoids corrupting our free list if the previous
 * WQE has already completed and been put on the free list when we
 * post the next WQE.
 */
static inline int *wqe_to_link(void *wqe)
{
	return (int *) (wqe + offsetof(struct mthca_next_seg, imm));
}

static void mthca_tavor_init_srq_context(struct mthca_dev *dev,
					 struct mthca_pd *pd,
					 struct mthca_srq *srq,
					 struct mthca_tavor_srq_context *context)
{
	memset(context, 0, sizeof *context);

	context->wqe_base_ds = cpu_to_be64(1 << (srq->wqe_shift - 4));
	context->state_pd    = cpu_to_be32(pd->pd_num);
	context->lkey        = cpu_to_be32(srq->mr.ibmr.lkey);

	if (pd->ibpd.uobject)
		context->uar =
			cpu_to_be32(to_mucontext(pd->ibpd.uobject->context)->uar.index);
	else
		context->uar = cpu_to_be32(dev->driver_uar.index);
}

static void mthca_arbel_init_srq_context(struct mthca_dev *dev,
					 struct mthca_pd *pd,
					 struct mthca_srq *srq,
					 struct mthca_arbel_srq_context *context)
{
	int logsize, max;

	memset(context, 0, sizeof *context);

	/*
	 * Put max in a temporary variable to work around gcc bug
	 * triggered by ilog2() on sparc64.
	 */
	max = srq->max;
	logsize = ilog2(max);
	context->state_logsize_srqn = cpu_to_be32(logsize << 24 | srq->srqn);
	context->lkey = cpu_to_be32(srq->mr.ibmr.lkey);
	context->db_index = cpu_to_be32(srq->db_index);
	context->logstride_usrpage = cpu_to_be32((srq->wqe_shift - 4) << 29);
	if (pd->ibpd.uobject)
		context->logstride_usrpage |=
			cpu_to_be32(to_mucontext(pd->ibpd.uobject->context)->uar.index);
	else
		context->logstride_usrpage |= cpu_to_be32(dev->driver_uar.index);
	context->eq_pd = cpu_to_be32(MTHCA_EQ_ASYNC << 24 | pd->pd_num);
}

static void mthca_free_srq_buf(struct mthca_dev *dev, struct mthca_srq *srq)
{
	mthca_buf_free(dev, srq->max << srq->wqe_shift, &srq->queue,
		       srq->is_direct, &srq->mr);
	kfree(srq->wrid);
}

static int mthca_alloc_srq_buf(struct mthca_dev *dev, struct mthca_pd *pd,
			       struct mthca_srq *srq)
{
	struct mthca_data_seg *scatter;
	void *wqe;
	int err;
	int i;

	if (pd->ibpd.uobject)
		return 0;

	srq->wrid = kmalloc(srq->max * sizeof (u64), GFP_KERNEL);
	if (!srq->wrid)
		return -ENOMEM;

	err = mthca_buf_alloc(dev, srq->max << srq->wqe_shift,
			      MTHCA_MAX_DIRECT_SRQ_SIZE,
			      &srq->queue, &srq->is_direct, pd, 1, &srq->mr);
	if (err) {
		kfree(srq->wrid);
		return err;
	}

	/*
	 * Now initialize the SRQ buffer so that all of the WQEs are
	 * linked into the list of free WQEs.  In addition, set the
	 * scatter list L_Keys to the sentry value of 0x100.
	 */
	for (i = 0; i < srq->max; ++i) {
		wqe = get_wqe(srq, i);

		*wqe_to_link(wqe) = i < srq->max - 1 ? i + 1 : -1;

		for (scatter = wqe + sizeof (struct mthca_next_seg);
		     (void *) scatter < wqe + (1 << srq->wqe_shift);
		     ++scatter)
			scatter->lkey = cpu_to_be32(MTHCA_INVAL_LKEY);
	}

	srq->last = get_wqe(srq, srq->max - 1);

	return 0;
}

int mthca_alloc_srq(struct mthca_dev *dev, struct mthca_pd *pd,
		    struct ib_srq_attr *attr, struct mthca_srq *srq)
{
	struct mthca_mailbox *mailbox;
	u8 status;
	int ds;
	int err;

	/* Sanity check SRQ size before proceeding */
	if (attr->max_wr  > dev->limits.max_srq_wqes ||
	    attr->max_sge > dev->limits.max_srq_sge)
		return -EINVAL;

	srq->max      = attr->max_wr;
	srq->max_gs   = attr->max_sge;
	srq->counter  = 0;

	if (mthca_is_memfree(dev))
		srq->max = roundup_pow_of_two(srq->max + 1);
	else
		srq->max = srq->max + 1;

	ds = max(64UL,
		 roundup_pow_of_two(sizeof (struct mthca_next_seg) +
				    srq->max_gs * sizeof (struct mthca_data_seg)));

	if (!mthca_is_memfree(dev) && (ds > dev->limits.max_desc_sz))
		return -EINVAL;

	srq->wqe_shift = ilog2(ds);

	srq->srqn = mthca_alloc(&dev->srq_table.alloc);
	if (srq->srqn == -1)
		return -ENOMEM;

	if (mthca_is_memfree(dev)) {
		err = mthca_table_get(dev, dev->srq_table.table, srq->srqn);
		if (err)
			goto err_out;

		if (!pd->ibpd.uobject) {
			srq->db_index = mthca_alloc_db(dev, MTHCA_DB_TYPE_SRQ,
						       srq->srqn, &srq->db);
			if (srq->db_index < 0) {
				err = -ENOMEM;
				goto err_out_icm;
			}
		}
	}

	mailbox = mthca_alloc_mailbox(dev, GFP_KERNEL);
	if (IS_ERR(mailbox)) {
		err = PTR_ERR(mailbox);
		goto err_out_db;
	}

	err = mthca_alloc_srq_buf(dev, pd, srq);
	if (err)
		goto err_out_mailbox;

	spin_lock_init(&srq->lock);
	srq->refcount = 1;
	init_waitqueue_head(&srq->wait);
	mutex_init(&srq->mutex);

	if (mthca_is_memfree(dev))
		mthca_arbel_init_srq_context(dev, pd, srq, mailbox->buf);
	else
		mthca_tavor_init_srq_context(dev, pd, srq, mailbox->buf);

	err = mthca_SW2HW_SRQ(dev, mailbox, srq->srqn, &status);

	if (err) {
		mthca_warn(dev, "SW2HW_SRQ failed (%d)\n", err);
		goto err_out_free_buf;
	}
	if (status) {
		mthca_warn(dev, "SW2HW_SRQ returned status 0x%02x\n",
			   status);
		err = -EINVAL;
		goto err_out_free_buf;
	}

	spin_lock_irq(&dev->srq_table.lock);
	if (mthca_array_set(&dev->srq_table.srq,
			    srq->srqn & (dev->limits.num_srqs - 1),
			    srq)) {
		spin_unlock_irq(&dev->srq_table.lock);
		goto err_out_free_srq;
	}
	spin_unlock_irq(&dev->srq_table.lock);

	mthca_free_mailbox(dev, mailbox);

	srq->first_free = 0;
	srq->last_free  = srq->max - 1;

	attr->max_wr    = srq->max - 1;
	attr->max_sge   = srq->max_gs;

	return 0;

err_out_free_srq:
	err = mthca_HW2SW_SRQ(dev, mailbox, srq->srqn, &status);
	if (err)
		mthca_warn(dev, "HW2SW_SRQ failed (%d)\n", err);
	else if (status)
		mthca_warn(dev, "HW2SW_SRQ returned status 0x%02x\n", status);

err_out_free_buf:
	if (!pd->ibpd.uobject)
		mthca_free_srq_buf(dev, srq);

err_out_mailbox:
	mthca_free_mailbox(dev, mailbox);

err_out_db:
	if (!pd->ibpd.uobject && mthca_is_memfree(dev))
		mthca_free_db(dev, MTHCA_DB_TYPE_SRQ, srq->db_index);

err_out_icm:
	mthca_table_put(dev, dev->srq_table.table, srq->srqn);

err_out:
	mthca_free(&dev->srq_table.alloc, srq->srqn);

	return err;
}

static inline int get_srq_refcount(struct mthca_dev *dev, struct mthca_srq *srq)
{
	int c;

	spin_lock_irq(&dev->srq_table.lock);
	c = srq->refcount;
	spin_unlock_irq(&dev->srq_table.lock);

	return c;
}

void mthca_free_srq(struct mthca_dev *dev, struct mthca_srq *srq)
{
	struct mthca_mailbox *mailbox;
	int err;
	u8 status;

	mailbox = mthca_alloc_mailbox(dev, GFP_KERNEL);
	if (IS_ERR(mailbox)) {
		mthca_warn(dev, "No memory for mailbox to free SRQ.\n");
		return;
	}

	err = mthca_HW2SW_SRQ(dev, mailbox, srq->srqn, &status);
	if (err)
		mthca_warn(dev, "HW2SW_SRQ failed (%d)\n", err);
	else if (status)
		mthca_warn(dev, "HW2SW_SRQ returned status 0x%02x\n", status);

	spin_lock_irq(&dev->srq_table.lock);
	mthca_array_clear(&dev->srq_table.srq,
			  srq->srqn & (dev->limits.num_srqs - 1));
	--srq->refcount;
	spin_unlock_irq(&dev->srq_table.lock);

	wait_event(srq->wait, !get_srq_refcount(dev, srq));

	if (!srq->ibsrq.uobject) {
		mthca_free_srq_buf(dev, srq);
		if (mthca_is_memfree(dev))
			mthca_free_db(dev, MTHCA_DB_TYPE_SRQ, srq->db_index);
	}

	mthca_table_put(dev, dev->srq_table.table, srq->srqn);
	mthca_free(&dev->srq_table.alloc, srq->srqn);
	mthca_free_mailbox(dev, mailbox);
}

int mthca_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
		     enum ib_srq_attr_mask attr_mask, struct ib_udata *udata)
{
	struct mthca_dev *dev = to_mdev(ibsrq->device);
	struct mthca_srq *srq = to_msrq(ibsrq);
	int ret;
	u8 status;

	/* We don't support resizing SRQs (yet?) */
	if (attr_mask & IB_SRQ_MAX_WR)
		return -EINVAL;

	if (attr_mask & IB_SRQ_LIMIT) {
		u32 max_wr = mthca_is_memfree(dev) ? srq->max - 1 : srq->max;
		if (attr->srq_limit > max_wr)
			return -EINVAL;

		mutex_lock(&srq->mutex);
		ret = mthca_ARM_SRQ(dev, srq->srqn, attr->srq_limit, &status);
		mutex_unlock(&srq->mutex);

		if (ret)
			return ret;
		if (status)
			return -EINVAL;
	}

	return 0;
}

int mthca_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *srq_attr)
{
	struct mthca_dev *dev = to_mdev(ibsrq->device);
	struct mthca_srq *srq = to_msrq(ibsrq);
	struct mthca_mailbox *mailbox;
	struct mthca_arbel_srq_context *arbel_ctx;
	struct mthca_tavor_srq_context *tavor_ctx;
	u8 status;
	int err;

	mailbox = mthca_alloc_mailbox(dev, GFP_KERNEL);
	if (IS_ERR(mailbox))
		return PTR_ERR(mailbox);

	err = mthca_QUERY_SRQ(dev, srq->srqn, mailbox, &status);
	if (err)
		goto out;

	if (mthca_is_memfree(dev)) {
		arbel_ctx = mailbox->buf;
		srq_attr->srq_limit = be16_to_cpu(arbel_ctx->limit_watermark);
	} else {
		tavor_ctx = mailbox->buf;
		srq_attr->srq_limit = be16_to_cpu(tavor_ctx->limit_watermark);
	}

	srq_attr->max_wr  = srq->max - 1;
	srq_attr->max_sge = srq->max_gs;

out:
	mthca_free_mailbox(dev, mailbox);

	return err;
}

void mthca_srq_event(struct mthca_dev *dev, u32 srqn,
		     enum ib_event_type event_type)
{
	struct mthca_srq *srq;
	struct ib_event event;

	spin_lock(&dev->srq_table.lock);
	srq = mthca_array_get(&dev->srq_table.srq, srqn & (dev->limits.num_srqs - 1));
	if (srq)
		++srq->refcount;
	spin_unlock(&dev->srq_table.lock);

	if (!srq) {
		mthca_warn(dev, "Async event for bogus SRQ %08x\n", srqn);
		return;
	}

	if (!srq->ibsrq.event_handler)
		goto out;

	event.device      = &dev->ib_dev;
	event.event       = event_type;
	event.element.srq = &srq->ibsrq;
	srq->ibsrq.event_handler(&event, srq->ibsrq.srq_context);

out:
	spin_lock(&dev->srq_table.lock);
	if (!--srq->refcount)
		wake_up(&srq->wait);
	spin_unlock(&dev->srq_table.lock);
}

/*
 * This function must be called with IRQs disabled.
 */
void mthca_free_srq_wqe(struct mthca_srq *srq, u32 wqe_addr)
{
	int ind;

	ind = wqe_addr >> srq->wqe_shift;

	spin_lock(&srq->lock);

	if (likely(srq->first_free >= 0))
		*wqe_to_link(get_wqe(srq, srq->last_free)) = ind;
	else
		srq->first_free = ind;

	*wqe_to_link(get_wqe(srq, ind)) = -1;
	srq->last_free = ind;

	spin_unlock(&srq->lock);
}

int mthca_tavor_post_srq_recv(struct ib_srq *ibsrq, struct ib_recv_wr *wr,
			      struct ib_recv_wr **bad_wr)
{
	struct mthca_dev *dev = to_mdev(ibsrq->device);
	struct mthca_srq *srq = to_msrq(ibsrq);
	__be32 doorbell[2];
	unsigned long flags;
	int err = 0;
	int first_ind;
	int ind;
	int next_ind;
	int nreq;
	int i;
	void *wqe;
	void *prev_wqe;

	spin_lock_irqsave(&srq->lock, flags);

	first_ind = srq->first_free;

	for (nreq = 0; wr; wr = wr->next) {
		ind = srq->first_free;

		if (ind < 0) {
			mthca_err(dev, "SRQ %06x full\n", srq->srqn);
			err = -ENOMEM;
			*bad_wr = wr;
			break;
		}

		wqe       = get_wqe(srq, ind);
		next_ind  = *wqe_to_link(wqe);

		if (next_ind < 0) {
			mthca_err(dev, "SRQ %06x full\n", srq->srqn);
			err = -ENOMEM;
			*bad_wr = wr;
			break;
		}

		prev_wqe  = srq->last;
		srq->last = wqe;

		((struct mthca_next_seg *) wqe)->nda_op = 0;
		((struct mthca_next_seg *) wqe)->ee_nds = 0;
		/* flags field will always remain 0 */

		wqe += sizeof (struct mthca_next_seg);

		if (unlikely(wr->num_sge > srq->max_gs)) {
			err = -EINVAL;
			*bad_wr = wr;
			srq->last = prev_wqe;
			break;
		}

		for (i = 0; i < wr->num_sge; ++i) {
			((struct mthca_data_seg *) wqe)->byte_count =
				cpu_to_be32(wr->sg_list[i].length);
			((struct mthca_data_seg *) wqe)->lkey =
				cpu_to_be32(wr->sg_list[i].lkey);
			((struct mthca_data_seg *) wqe)->addr =
				cpu_to_be64(wr->sg_list[i].addr);
			wqe += sizeof (struct mthca_data_seg);
		}

		if (i < srq->max_gs) {
			((struct mthca_data_seg *) wqe)->byte_count = 0;
			((struct mthca_data_seg *) wqe)->lkey = cpu_to_be32(MTHCA_INVAL_LKEY);
			((struct mthca_data_seg *) wqe)->addr = 0;
		}

		((struct mthca_next_seg *) prev_wqe)->nda_op =
			cpu_to_be32((ind << srq->wqe_shift) | 1);
		wmb();
		((struct mthca_next_seg *) prev_wqe)->ee_nds =
			cpu_to_be32(MTHCA_NEXT_DBD);

		srq->wrid[ind]  = wr->wr_id;
		srq->first_free = next_ind;

		++nreq;
		if (unlikely(nreq == MTHCA_TAVOR_MAX_WQES_PER_RECV_DB)) {
			nreq = 0;

			doorbell[0] = cpu_to_be32(first_ind << srq->wqe_shift);
			doorbell[1] = cpu_to_be32(srq->srqn << 8);

			/*
			 * Make sure that descriptors are written
			 * before doorbell is rung.
			 */
			wmb();

			mthca_write64(doorbell,
				      dev->kar + MTHCA_RECEIVE_DOORBELL,
				      MTHCA_GET_DOORBELL_LOCK(&dev->doorbell_lock));

			first_ind = srq->first_free;
		}
	}

	if (likely(nreq)) {
		doorbell[0] = cpu_to_be32(first_ind << srq->wqe_shift);
		doorbell[1] = cpu_to_be32((srq->srqn << 8) | nreq);

		/*
		 * Make sure that descriptors are written before
		 * doorbell is rung.
		 */
		wmb();

		mthca_write64(doorbell,
			      dev->kar + MTHCA_RECEIVE_DOORBELL,
			      MTHCA_GET_DOORBELL_LOCK(&dev->doorbell_lock));
	}

	/*
	 * Make sure doorbells don't leak out of SRQ spinlock and
	 * reach the HCA out of order:
	 */
	mmiowb();

	spin_unlock_irqrestore(&srq->lock, flags);
	return err;
}

int mthca_arbel_post_srq_recv(struct ib_srq *ibsrq, struct ib_recv_wr *wr,
			      struct ib_recv_wr **bad_wr)
{
	struct mthca_dev *dev = to_mdev(ibsrq->device);
	struct mthca_srq *srq = to_msrq(ibsrq);
	unsigned long flags;
	int err = 0;
	int ind;
	int next_ind;
	int nreq;
	int i;
	void *wqe;

	spin_lock_irqsave(&srq->lock, flags);

	for (nreq = 0; wr; ++nreq, wr = wr->next) {
		ind = srq->first_free;

		if (ind < 0) {
			mthca_err(dev, "SRQ %06x full\n", srq->srqn);
			err = -ENOMEM;
			*bad_wr = wr;
			break;
		}

		wqe       = get_wqe(srq, ind);
		next_ind  = *wqe_to_link(wqe);

		if (next_ind < 0) {
			mthca_err(dev, "SRQ %06x full\n", srq->srqn);
			err = -ENOMEM;
			*bad_wr = wr;
			break;
		}

		((struct mthca_next_seg *) wqe)->nda_op =
			cpu_to_be32((next_ind << srq->wqe_shift) | 1);
		((struct mthca_next_seg *) wqe)->ee_nds = 0;
		/* flags field will always remain 0 */

		wqe += sizeof (struct mthca_next_seg);

		if (unlikely(wr->num_sge > srq->max_gs)) {
			err = -EINVAL;
			*bad_wr = wr;
			break;
		}

		for (i = 0; i < wr->num_sge; ++i) {
			((struct mthca_data_seg *) wqe)->byte_count =
				cpu_to_be32(wr->sg_list[i].length);
			((struct mthca_data_seg *) wqe)->lkey =
				cpu_to_be32(wr->sg_list[i].lkey);
			((struct mthca_data_seg *) wqe)->addr =
				cpu_to_be64(wr->sg_list[i].addr);
			wqe += sizeof (struct mthca_data_seg);
		}

		if (i < srq->max_gs) {
			((struct mthca_data_seg *) wqe)->byte_count = 0;
			((struct mthca_data_seg *) wqe)->lkey = cpu_to_be32(MTHCA_INVAL_LKEY);
			((struct mthca_data_seg *) wqe)->addr = 0;
		}

		srq->wrid[ind]  = wr->wr_id;
		srq->first_free = next_ind;
	}

	if (likely(nreq)) {
		srq->counter += nreq;

		/*
		 * Make sure that descriptors are written before
		 * we write doorbell record.
		 */
		wmb();
		*srq->db = cpu_to_be32(srq->counter);
	}

	spin_unlock_irqrestore(&srq->lock, flags);
	return err;
}

int mthca_max_srq_sge(struct mthca_dev *dev)
{
	if (mthca_is_memfree(dev))
		return dev->limits.max_sg;

	/*
	 * SRQ allocations are based on powers of 2 for Tavor,
	 * (although they only need to be multiples of 16 bytes).
	 *
	 * Therefore, we need to base the max number of sg entries on
	 * the largest power of 2 descriptor size that is <= to the
	 * actual max WQE descriptor size, rather than return the
	 * max_sg value given by the firmware (which is based on WQE
	 * sizes as multiples of 16, not powers of 2).
	 *
	 * If SRQ implementation is changed for Tavor to be based on
	 * multiples of 16, the calculation below can be deleted and
	 * the FW max_sg value returned.
	 */
	return min_t(int, dev->limits.max_sg,
		     ((1 << (fls(dev->limits.max_desc_sz) - 1)) -
		      sizeof (struct mthca_next_seg)) /
		     sizeof (struct mthca_data_seg));
}

int mthca_init_srq_table(struct mthca_dev *dev)
{
	int err;

	if (!(dev->mthca_flags & MTHCA_FLAG_SRQ))
		return 0;

	spin_lock_init(&dev->srq_table.lock);

	err = mthca_alloc_init(&dev->srq_table.alloc,
			       dev->limits.num_srqs,
			       dev->limits.num_srqs - 1,
			       dev->limits.reserved_srqs);
	if (err)
		return err;

	err = mthca_array_init(&dev->srq_table.srq,
			       dev->limits.num_srqs);
	if (err)
		mthca_alloc_cleanup(&dev->srq_table.alloc);

	return err;
}

void mthca_cleanup_srq_table(struct mthca_dev *dev)
{
	if (!(dev->mthca_flags & MTHCA_FLAG_SRQ))
		return;

	mthca_array_cleanup(&dev->srq_table.srq, dev->limits.num_srqs);
	mthca_alloc_cleanup(&dev->srq_table.alloc);
}