summaryrefslogtreecommitdiffstats
path: root/drivers/infiniband/hw/qib/qib_sdma.c
blob: 8f8d6173665682a45814802a01ef137d299a45cf (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
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
/*
 * Copyright (c) 2012 Intel Corporation. All rights reserved.
 * Copyright (c) 2007 - 2012 QLogic Corporation. 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 <linux/netdevice.h>
#include <linux/moduleparam.h>

#include "qib.h"
#include "qib_common.h"

/* default pio off, sdma on */
static ushort sdma_descq_cnt = 256;
module_param_named(sdma_descq_cnt, sdma_descq_cnt, ushort, S_IRUGO);
MODULE_PARM_DESC(sdma_descq_cnt, "Number of SDMA descq entries");

/*
 * Bits defined in the send DMA descriptor.
 */
#define SDMA_DESC_LAST          (1ULL << 11)
#define SDMA_DESC_FIRST         (1ULL << 12)
#define SDMA_DESC_DMA_HEAD      (1ULL << 13)
#define SDMA_DESC_USE_LARGE_BUF (1ULL << 14)
#define SDMA_DESC_INTR          (1ULL << 15)
#define SDMA_DESC_COUNT_LSB     16
#define SDMA_DESC_GEN_LSB       30

/* declare all statics here rather than keep sorting */
static int alloc_sdma(struct qib_pportdata *);
static void sdma_complete(struct kref *);
static void sdma_finalput(struct qib_sdma_state *);
static void sdma_get(struct qib_sdma_state *);
static void sdma_put(struct qib_sdma_state *);
static void sdma_set_state(struct qib_pportdata *, enum qib_sdma_states);
static void sdma_start_sw_clean_up(struct qib_pportdata *);
static void sdma_sw_clean_up_task(unsigned long);
static void unmap_desc(struct qib_pportdata *, unsigned);

static void sdma_get(struct qib_sdma_state *ss)
{
	kref_get(&ss->kref);
}

static void sdma_complete(struct kref *kref)
{
	struct qib_sdma_state *ss =
		container_of(kref, struct qib_sdma_state, kref);

	complete(&ss->comp);
}

static void sdma_put(struct qib_sdma_state *ss)
{
	kref_put(&ss->kref, sdma_complete);
}

static void sdma_finalput(struct qib_sdma_state *ss)
{
	sdma_put(ss);
	wait_for_completion(&ss->comp);
}

/*
 * Complete all the sdma requests on the active list, in the correct
 * order, and with appropriate processing.   Called when cleaning up
 * after sdma shutdown, and when new sdma requests are submitted for
 * a link that is down.   This matches what is done for requests
 * that complete normally, it's just the full list.
 *
 * Must be called with sdma_lock held
 */
static void clear_sdma_activelist(struct qib_pportdata *ppd)
{
	struct qib_sdma_txreq *txp, *txp_next;

	list_for_each_entry_safe(txp, txp_next, &ppd->sdma_activelist, list) {
		list_del_init(&txp->list);
		if (txp->flags & QIB_SDMA_TXREQ_F_FREEDESC) {
			unsigned idx;

			idx = txp->start_idx;
			while (idx != txp->next_descq_idx) {
				unmap_desc(ppd, idx);
				if (++idx == ppd->sdma_descq_cnt)
					idx = 0;
			}
		}
		if (txp->callback)
			(*txp->callback)(txp, QIB_SDMA_TXREQ_S_ABORTED);
	}
}

static void sdma_sw_clean_up_task(unsigned long opaque)
{
	struct qib_pportdata *ppd = (struct qib_pportdata *) opaque;
	unsigned long flags;

	spin_lock_irqsave(&ppd->sdma_lock, flags);

	/*
	 * At this point, the following should always be true:
	 * - We are halted, so no more descriptors are getting retired.
	 * - We are not running, so no one is submitting new work.
	 * - Only we can send the e40_sw_cleaned, so we can't start
	 *   running again until we say so.  So, the active list and
	 *   descq are ours to play with.
	 */

	/* Process all retired requests. */
	qib_sdma_make_progress(ppd);

	clear_sdma_activelist(ppd);

	/*
	 * Resync count of added and removed.  It is VERY important that
	 * sdma_descq_removed NEVER decrement - user_sdma depends on it.
	 */
	ppd->sdma_descq_removed = ppd->sdma_descq_added;

	/*
	 * Reset our notion of head and tail.
	 * Note that the HW registers will be reset when switching states
	 * due to calling __qib_sdma_process_event() below.
	 */
	ppd->sdma_descq_tail = 0;
	ppd->sdma_descq_head = 0;
	ppd->sdma_head_dma[0] = 0;
	ppd->sdma_generation = 0;

	__qib_sdma_process_event(ppd, qib_sdma_event_e40_sw_cleaned);

	spin_unlock_irqrestore(&ppd->sdma_lock, flags);
}

/*
 * This is called when changing to state qib_sdma_state_s10_hw_start_up_wait
 * as a result of send buffer errors or send DMA descriptor errors.
 * We want to disarm the buffers in these cases.
 */
static void sdma_hw_start_up(struct qib_pportdata *ppd)
{
	struct qib_sdma_state *ss = &ppd->sdma_state;
	unsigned bufno;

	for (bufno = ss->first_sendbuf; bufno < ss->last_sendbuf; ++bufno)
		ppd->dd->f_sendctrl(ppd, QIB_SENDCTRL_DISARM_BUF(bufno));

	ppd->dd->f_sdma_hw_start_up(ppd);
}

static void sdma_sw_tear_down(struct qib_pportdata *ppd)
{
	struct qib_sdma_state *ss = &ppd->sdma_state;

	/* Releasing this reference means the state machine has stopped. */
	sdma_put(ss);
}

static void sdma_start_sw_clean_up(struct qib_pportdata *ppd)
{
	tasklet_hi_schedule(&ppd->sdma_sw_clean_up_task);
}

static void sdma_set_state(struct qib_pportdata *ppd,
	enum qib_sdma_states next_state)
{
	struct qib_sdma_state *ss = &ppd->sdma_state;
	struct sdma_set_state_action *action = ss->set_state_action;
	unsigned op = 0;

	/* debugging bookkeeping */
	ss->previous_state = ss->current_state;
	ss->previous_op = ss->current_op;

	ss->current_state = next_state;

	if (action[next_state].op_enable)
		op |= QIB_SDMA_SENDCTRL_OP_ENABLE;

	if (action[next_state].op_intenable)
		op |= QIB_SDMA_SENDCTRL_OP_INTENABLE;

	if (action[next_state].op_halt)
		op |= QIB_SDMA_SENDCTRL_OP_HALT;

	if (action[next_state].op_drain)
		op |= QIB_SDMA_SENDCTRL_OP_DRAIN;

	if (action[next_state].go_s99_running_tofalse)
		ss->go_s99_running = 0;

	if (action[next_state].go_s99_running_totrue)
		ss->go_s99_running = 1;

	ss->current_op = op;

	ppd->dd->f_sdma_sendctrl(ppd, ss->current_op);
}

static void unmap_desc(struct qib_pportdata *ppd, unsigned head)
{
	__le64 *descqp = &ppd->sdma_descq[head].qw[0];
	u64 desc[2];
	dma_addr_t addr;
	size_t len;

	desc[0] = le64_to_cpu(descqp[0]);
	desc[1] = le64_to_cpu(descqp[1]);

	addr = (desc[1] << 32) | (desc[0] >> 32);
	len = (desc[0] >> 14) & (0x7ffULL << 2);
	dma_unmap_single(&ppd->dd->pcidev->dev, addr, len, DMA_TO_DEVICE);
}

static int alloc_sdma(struct qib_pportdata *ppd)
{
	ppd->sdma_descq_cnt = sdma_descq_cnt;
	if (!ppd->sdma_descq_cnt)
		ppd->sdma_descq_cnt = 256;

	/* Allocate memory for SendDMA descriptor FIFO */
	ppd->sdma_descq = dma_alloc_coherent(&ppd->dd->pcidev->dev,
		ppd->sdma_descq_cnt * sizeof(u64[2]), &ppd->sdma_descq_phys,
		GFP_KERNEL);

	if (!ppd->sdma_descq) {
		qib_dev_err(ppd->dd,
			"failed to allocate SendDMA descriptor FIFO memory\n");
		goto bail;
	}

	/* Allocate memory for DMA of head register to memory */
	ppd->sdma_head_dma = dma_alloc_coherent(&ppd->dd->pcidev->dev,
		PAGE_SIZE, &ppd->sdma_head_phys, GFP_KERNEL);
	if (!ppd->sdma_head_dma) {
		qib_dev_err(ppd->dd,
			"failed to allocate SendDMA head memory\n");
		goto cleanup_descq;
	}
	ppd->sdma_head_dma[0] = 0;
	return 0;

cleanup_descq:
	dma_free_coherent(&ppd->dd->pcidev->dev,
		ppd->sdma_descq_cnt * sizeof(u64[2]), (void *)ppd->sdma_descq,
		ppd->sdma_descq_phys);
	ppd->sdma_descq = NULL;
	ppd->sdma_descq_phys = 0;
bail:
	ppd->sdma_descq_cnt = 0;
	return -ENOMEM;
}

static void free_sdma(struct qib_pportdata *ppd)
{
	struct qib_devdata *dd = ppd->dd;

	if (ppd->sdma_head_dma) {
		dma_free_coherent(&dd->pcidev->dev, PAGE_SIZE,
				  (void *)ppd->sdma_head_dma,
				  ppd->sdma_head_phys);
		ppd->sdma_head_dma = NULL;
		ppd->sdma_head_phys = 0;
	}

	if (ppd->sdma_descq) {
		dma_free_coherent(&dd->pcidev->dev,
				  ppd->sdma_descq_cnt * sizeof(u64[2]),
				  ppd->sdma_descq, ppd->sdma_descq_phys);
		ppd->sdma_descq = NULL;
		ppd->sdma_descq_phys = 0;
	}
}

static inline void make_sdma_desc(struct qib_pportdata *ppd,
				  u64 *sdmadesc, u64 addr, u64 dwlen,
				  u64 dwoffset)
{

	WARN_ON(addr & 3);
	/* SDmaPhyAddr[47:32] */
	sdmadesc[1] = addr >> 32;
	/* SDmaPhyAddr[31:0] */
	sdmadesc[0] = (addr & 0xfffffffcULL) << 32;
	/* SDmaGeneration[1:0] */
	sdmadesc[0] |= (ppd->sdma_generation & 3ULL) <<
		SDMA_DESC_GEN_LSB;
	/* SDmaDwordCount[10:0] */
	sdmadesc[0] |= (dwlen & 0x7ffULL) << SDMA_DESC_COUNT_LSB;
	/* SDmaBufOffset[12:2] */
	sdmadesc[0] |= dwoffset & 0x7ffULL;
}

/* sdma_lock must be held */
int qib_sdma_make_progress(struct qib_pportdata *ppd)
{
	struct list_head *lp = NULL;
	struct qib_sdma_txreq *txp = NULL;
	struct qib_devdata *dd = ppd->dd;
	int progress = 0;
	u16 hwhead;
	u16 idx = 0;

	hwhead = dd->f_sdma_gethead(ppd);

	/* The reason for some of the complexity of this code is that
	 * not all descriptors have corresponding txps.  So, we have to
	 * be able to skip over descs until we wander into the range of
	 * the next txp on the list.
	 */

	if (!list_empty(&ppd->sdma_activelist)) {
		lp = ppd->sdma_activelist.next;
		txp = list_entry(lp, struct qib_sdma_txreq, list);
		idx = txp->start_idx;
	}

	while (ppd->sdma_descq_head != hwhead) {
		/* if desc is part of this txp, unmap if needed */
		if (txp && (txp->flags & QIB_SDMA_TXREQ_F_FREEDESC) &&
		    (idx == ppd->sdma_descq_head)) {
			unmap_desc(ppd, ppd->sdma_descq_head);
			if (++idx == ppd->sdma_descq_cnt)
				idx = 0;
		}

		/* increment dequed desc count */
		ppd->sdma_descq_removed++;

		/* advance head, wrap if needed */
		if (++ppd->sdma_descq_head == ppd->sdma_descq_cnt)
			ppd->sdma_descq_head = 0;

		/* if now past this txp's descs, do the callback */
		if (txp && txp->next_descq_idx == ppd->sdma_descq_head) {
			/* remove from active list */
			list_del_init(&txp->list);
			if (txp->callback)
				(*txp->callback)(txp, QIB_SDMA_TXREQ_S_OK);
			/* see if there is another txp */
			if (list_empty(&ppd->sdma_activelist))
				txp = NULL;
			else {
				lp = ppd->sdma_activelist.next;
				txp = list_entry(lp, struct qib_sdma_txreq,
					list);
				idx = txp->start_idx;
			}
		}
		progress = 1;
	}
	if (progress)
		qib_verbs_sdma_desc_avail(ppd, qib_sdma_descq_freecnt(ppd));
	return progress;
}

/*
 * This is called from interrupt context.
 */
void qib_sdma_intr(struct qib_pportdata *ppd)
{
	unsigned long flags;

	spin_lock_irqsave(&ppd->sdma_lock, flags);

	__qib_sdma_intr(ppd);

	spin_unlock_irqrestore(&ppd->sdma_lock, flags);
}

void __qib_sdma_intr(struct qib_pportdata *ppd)
{
	if (__qib_sdma_running(ppd)) {
		qib_sdma_make_progress(ppd);
		if (!list_empty(&ppd->sdma_userpending))
			qib_user_sdma_send_desc(ppd, &ppd->sdma_userpending);
	}
}

int qib_setup_sdma(struct qib_pportdata *ppd)
{
	struct qib_devdata *dd = ppd->dd;
	unsigned long flags;
	int ret = 0;

	ret = alloc_sdma(ppd);
	if (ret)
		goto bail;

	/* set consistent sdma state */
	ppd->dd->f_sdma_init_early(ppd);
	spin_lock_irqsave(&ppd->sdma_lock, flags);
	sdma_set_state(ppd, qib_sdma_state_s00_hw_down);
	spin_unlock_irqrestore(&ppd->sdma_lock, flags);

	/* set up reference counting */
	kref_init(&ppd->sdma_state.kref);
	init_completion(&ppd->sdma_state.comp);

	ppd->sdma_generation = 0;
	ppd->sdma_descq_head = 0;
	ppd->sdma_descq_removed = 0;
	ppd->sdma_descq_added = 0;

	ppd->sdma_intrequest = 0;
	INIT_LIST_HEAD(&ppd->sdma_userpending);

	INIT_LIST_HEAD(&ppd->sdma_activelist);

	tasklet_init(&ppd->sdma_sw_clean_up_task, sdma_sw_clean_up_task,
		(unsigned long)ppd);

	ret = dd->f_init_sdma_regs(ppd);
	if (ret)
		goto bail_alloc;

	qib_sdma_process_event(ppd, qib_sdma_event_e10_go_hw_start);

	return 0;

bail_alloc:
	qib_teardown_sdma(ppd);
bail:
	return ret;
}

void qib_teardown_sdma(struct qib_pportdata *ppd)
{
	qib_sdma_process_event(ppd, qib_sdma_event_e00_go_hw_down);

	/*
	 * This waits for the state machine to exit so it is not
	 * necessary to kill the sdma_sw_clean_up_task to make sure
	 * it is not running.
	 */
	sdma_finalput(&ppd->sdma_state);

	free_sdma(ppd);
}

int qib_sdma_running(struct qib_pportdata *ppd)
{
	unsigned long flags;
	int ret;

	spin_lock_irqsave(&ppd->sdma_lock, flags);
	ret = __qib_sdma_running(ppd);
	spin_unlock_irqrestore(&ppd->sdma_lock, flags);

	return ret;
}

/*
 * Complete a request when sdma not running; likely only request
 * but to simplify the code, always queue it, then process the full
 * activelist.  We process the entire list to ensure that this particular
 * request does get it's callback, but in the correct order.
 * Must be called with sdma_lock held
 */
static void complete_sdma_err_req(struct qib_pportdata *ppd,
				  struct qib_verbs_txreq *tx)
{
	struct qib_qp_priv *priv = tx->qp->priv;

	atomic_inc(&priv->s_dma_busy);
	/* no sdma descriptors, so no unmap_desc */
	tx->txreq.start_idx = 0;
	tx->txreq.next_descq_idx = 0;
	list_add_tail(&tx->txreq.list, &ppd->sdma_activelist);
	clear_sdma_activelist(ppd);
}

/*
 * This function queues one IB packet onto the send DMA queue per call.
 * The caller is responsible for checking:
 * 1) The number of send DMA descriptor entries is less than the size of
 *    the descriptor queue.
 * 2) The IB SGE addresses and lengths are 32-bit aligned
 *    (except possibly the last SGE's length)
 * 3) The SGE addresses are suitable for passing to dma_map_single().
 */
int qib_sdma_verbs_send(struct qib_pportdata *ppd,
			struct rvt_sge_state *ss, u32 dwords,
			struct qib_verbs_txreq *tx)
{
	unsigned long flags;
	struct rvt_sge *sge;
	struct rvt_qp *qp;
	int ret = 0;
	u16 tail;
	__le64 *descqp;
	u64 sdmadesc[2];
	u32 dwoffset;
	dma_addr_t addr;
	struct qib_qp_priv *priv;

	spin_lock_irqsave(&ppd->sdma_lock, flags);

retry:
	if (unlikely(!__qib_sdma_running(ppd))) {
		complete_sdma_err_req(ppd, tx);
		goto unlock;
	}

	if (tx->txreq.sg_count > qib_sdma_descq_freecnt(ppd)) {
		if (qib_sdma_make_progress(ppd))
			goto retry;
		if (ppd->dd->flags & QIB_HAS_SDMA_TIMEOUT)
			ppd->dd->f_sdma_set_desc_cnt(ppd,
					ppd->sdma_descq_cnt / 2);
		goto busy;
	}

	dwoffset = tx->hdr_dwords;
	make_sdma_desc(ppd, sdmadesc, (u64) tx->txreq.addr, dwoffset, 0);

	sdmadesc[0] |= SDMA_DESC_FIRST;
	if (tx->txreq.flags & QIB_SDMA_TXREQ_F_USELARGEBUF)
		sdmadesc[0] |= SDMA_DESC_USE_LARGE_BUF;

	/* write to the descq */
	tail = ppd->sdma_descq_tail;
	descqp = &ppd->sdma_descq[tail].qw[0];
	*descqp++ = cpu_to_le64(sdmadesc[0]);
	*descqp++ = cpu_to_le64(sdmadesc[1]);

	/* increment the tail */
	if (++tail == ppd->sdma_descq_cnt) {
		tail = 0;
		descqp = &ppd->sdma_descq[0].qw[0];
		++ppd->sdma_generation;
	}

	tx->txreq.start_idx = tail;

	sge = &ss->sge;
	while (dwords) {
		u32 dw;
		u32 len = rvt_get_sge_length(sge, dwords << 2);

		dw = (len + 3) >> 2;
		addr = dma_map_single(&ppd->dd->pcidev->dev, sge->vaddr,
				      dw << 2, DMA_TO_DEVICE);
		if (dma_mapping_error(&ppd->dd->pcidev->dev, addr)) {
			ret = -ENOMEM;
			goto unmap;
		}
		sdmadesc[0] = 0;
		make_sdma_desc(ppd, sdmadesc, (u64) addr, dw, dwoffset);
		/* SDmaUseLargeBuf has to be set in every descriptor */
		if (tx->txreq.flags & QIB_SDMA_TXREQ_F_USELARGEBUF)
			sdmadesc[0] |= SDMA_DESC_USE_LARGE_BUF;
		/* write to the descq */
		*descqp++ = cpu_to_le64(sdmadesc[0]);
		*descqp++ = cpu_to_le64(sdmadesc[1]);

		/* increment the tail */
		if (++tail == ppd->sdma_descq_cnt) {
			tail = 0;
			descqp = &ppd->sdma_descq[0].qw[0];
			++ppd->sdma_generation;
		}
		rvt_update_sge(ss, len, false);
		dwoffset += dw;
		dwords -= dw;
	}

	if (!tail)
		descqp = &ppd->sdma_descq[ppd->sdma_descq_cnt].qw[0];
	descqp -= 2;
	descqp[0] |= cpu_to_le64(SDMA_DESC_LAST);
	if (tx->txreq.flags & QIB_SDMA_TXREQ_F_HEADTOHOST)
		descqp[0] |= cpu_to_le64(SDMA_DESC_DMA_HEAD);
	if (tx->txreq.flags & QIB_SDMA_TXREQ_F_INTREQ)
		descqp[0] |= cpu_to_le64(SDMA_DESC_INTR);
	priv = tx->qp->priv;
	atomic_inc(&priv->s_dma_busy);
	tx->txreq.next_descq_idx = tail;
	ppd->dd->f_sdma_update_tail(ppd, tail);
	ppd->sdma_descq_added += tx->txreq.sg_count;
	list_add_tail(&tx->txreq.list, &ppd->sdma_activelist);
	goto unlock;

unmap:
	for (;;) {
		if (!tail)
			tail = ppd->sdma_descq_cnt - 1;
		else
			tail--;
		if (tail == ppd->sdma_descq_tail)
			break;
		unmap_desc(ppd, tail);
	}
	qp = tx->qp;
	priv = qp->priv;
	qib_put_txreq(tx);
	spin_lock(&qp->r_lock);
	spin_lock(&qp->s_lock);
	if (qp->ibqp.qp_type == IB_QPT_RC) {
		/* XXX what about error sending RDMA read responses? */
		if (ib_rvt_state_ops[qp->state] & RVT_PROCESS_RECV_OK)
			rvt_error_qp(qp, IB_WC_GENERAL_ERR);
	} else if (qp->s_wqe)
		rvt_send_complete(qp, qp->s_wqe, IB_WC_GENERAL_ERR);
	spin_unlock(&qp->s_lock);
	spin_unlock(&qp->r_lock);
	/* return zero to process the next send work request */
	goto unlock;

busy:
	qp = tx->qp;
	priv = qp->priv;
	spin_lock(&qp->s_lock);
	if (ib_rvt_state_ops[qp->state] & RVT_PROCESS_RECV_OK) {
		struct qib_ibdev *dev;

		/*
		 * If we couldn't queue the DMA request, save the info
		 * and try again later rather than destroying the
		 * buffer and undoing the side effects of the copy.
		 */
		tx->ss = ss;
		tx->dwords = dwords;
		priv->s_tx = tx;
		dev = &ppd->dd->verbs_dev;
		spin_lock(&dev->rdi.pending_lock);
		if (list_empty(&priv->iowait)) {
			struct qib_ibport *ibp;

			ibp = &ppd->ibport_data;
			ibp->rvp.n_dmawait++;
			qp->s_flags |= RVT_S_WAIT_DMA_DESC;
			list_add_tail(&priv->iowait, &dev->dmawait);
		}
		spin_unlock(&dev->rdi.pending_lock);
		qp->s_flags &= ~RVT_S_BUSY;
		spin_unlock(&qp->s_lock);
		ret = -EBUSY;
	} else {
		spin_unlock(&qp->s_lock);
		qib_put_txreq(tx);
	}
unlock:
	spin_unlock_irqrestore(&ppd->sdma_lock, flags);
	return ret;
}

/*
 * sdma_lock should be acquired before calling this routine
 */
void dump_sdma_state(struct qib_pportdata *ppd)
{
	struct qib_sdma_desc *descq;
	struct qib_sdma_txreq *txp, *txpnext;
	__le64 *descqp;
	u64 desc[2];
	u64 addr;
	u16 gen, dwlen, dwoffset;
	u16 head, tail, cnt;

	head = ppd->sdma_descq_head;
	tail = ppd->sdma_descq_tail;
	cnt = qib_sdma_descq_freecnt(ppd);
	descq = ppd->sdma_descq;

	qib_dev_porterr(ppd->dd, ppd->port,
		"SDMA ppd->sdma_descq_head: %u\n", head);
	qib_dev_porterr(ppd->dd, ppd->port,
		"SDMA ppd->sdma_descq_tail: %u\n", tail);
	qib_dev_porterr(ppd->dd, ppd->port,
		"SDMA sdma_descq_freecnt: %u\n", cnt);

	/* print info for each entry in the descriptor queue */
	while (head != tail) {
		char flags[6] = { 'x', 'x', 'x', 'x', 'x', 0 };

		descqp = &descq[head].qw[0];
		desc[0] = le64_to_cpu(descqp[0]);
		desc[1] = le64_to_cpu(descqp[1]);
		flags[0] = (desc[0] & 1<<15) ? 'I' : '-';
		flags[1] = (desc[0] & 1<<14) ? 'L' : 'S';
		flags[2] = (desc[0] & 1<<13) ? 'H' : '-';
		flags[3] = (desc[0] & 1<<12) ? 'F' : '-';
		flags[4] = (desc[0] & 1<<11) ? 'L' : '-';
		addr = (desc[1] << 32) | ((desc[0] >> 32) & 0xfffffffcULL);
		gen = (desc[0] >> 30) & 3ULL;
		dwlen = (desc[0] >> 14) & (0x7ffULL << 2);
		dwoffset = (desc[0] & 0x7ffULL) << 2;
		qib_dev_porterr(ppd->dd, ppd->port,
			"SDMA sdmadesc[%u]: flags:%s addr:0x%016llx gen:%u len:%u bytes offset:%u bytes\n",
			 head, flags, addr, gen, dwlen, dwoffset);
		if (++head == ppd->sdma_descq_cnt)
			head = 0;
	}

	/* print dma descriptor indices from the TX requests */
	list_for_each_entry_safe(txp, txpnext, &ppd->sdma_activelist,
				 list)
		qib_dev_porterr(ppd->dd, ppd->port,
			"SDMA txp->start_idx: %u txp->next_descq_idx: %u\n",
			txp->start_idx, txp->next_descq_idx);
}

void qib_sdma_process_event(struct qib_pportdata *ppd,
	enum qib_sdma_events event)
{
	unsigned long flags;

	spin_lock_irqsave(&ppd->sdma_lock, flags);

	__qib_sdma_process_event(ppd, event);

	if (ppd->sdma_state.current_state == qib_sdma_state_s99_running)
		qib_verbs_sdma_desc_avail(ppd, qib_sdma_descq_freecnt(ppd));

	spin_unlock_irqrestore(&ppd->sdma_lock, flags);
}

void __qib_sdma_process_event(struct qib_pportdata *ppd,
	enum qib_sdma_events event)
{
	struct qib_sdma_state *ss = &ppd->sdma_state;

	switch (ss->current_state) {
	case qib_sdma_state_s00_hw_down:
		switch (event) {
		case qib_sdma_event_e00_go_hw_down:
			break;
		case qib_sdma_event_e30_go_running:
			/*
			 * If down, but running requested (usually result
			 * of link up, then we need to start up.
			 * This can happen when hw down is requested while
			 * bringing the link up with traffic active on
			 * 7220, e.g. */
			ss->go_s99_running = 1;
			fallthrough;	/* and start dma engine */
		case qib_sdma_event_e10_go_hw_start:
			/* This reference means the state machine is started */
			sdma_get(&ppd->sdma_state);
			sdma_set_state(ppd,
				       qib_sdma_state_s10_hw_start_up_wait);
			break;
		case qib_sdma_event_e20_hw_started:
			break;
		case qib_sdma_event_e40_sw_cleaned:
			sdma_sw_tear_down(ppd);
			break;
		case qib_sdma_event_e50_hw_cleaned:
			break;
		case qib_sdma_event_e60_hw_halted:
			break;
		case qib_sdma_event_e70_go_idle:
			break;
		case qib_sdma_event_e7220_err_halted:
			break;
		case qib_sdma_event_e7322_err_halted:
			break;
		case qib_sdma_event_e90_timer_tick:
			break;
		}
		break;

	case qib_sdma_state_s10_hw_start_up_wait:
		switch (event) {
		case qib_sdma_event_e00_go_hw_down:
			sdma_set_state(ppd, qib_sdma_state_s00_hw_down);
			sdma_sw_tear_down(ppd);
			break;
		case qib_sdma_event_e10_go_hw_start:
			break;
		case qib_sdma_event_e20_hw_started:
			sdma_set_state(ppd, ss->go_s99_running ?
				       qib_sdma_state_s99_running :
				       qib_sdma_state_s20_idle);
			break;
		case qib_sdma_event_e30_go_running:
			ss->go_s99_running = 1;
			break;
		case qib_sdma_event_e40_sw_cleaned:
			break;
		case qib_sdma_event_e50_hw_cleaned:
			break;
		case qib_sdma_event_e60_hw_halted:
			break;
		case qib_sdma_event_e70_go_idle:
			ss->go_s99_running = 0;
			break;
		case qib_sdma_event_e7220_err_halted:
			break;
		case qib_sdma_event_e7322_err_halted:
			break;
		case qib_sdma_event_e90_timer_tick:
			break;
		}
		break;

	case qib_sdma_state_s20_idle:
		switch (event) {
		case qib_sdma_event_e00_go_hw_down:
			sdma_set_state(ppd, qib_sdma_state_s00_hw_down);
			sdma_sw_tear_down(ppd);
			break;
		case qib_sdma_event_e10_go_hw_start:
			break;
		case qib_sdma_event_e20_hw_started:
			break;
		case qib_sdma_event_e30_go_running:
			sdma_set_state(ppd, qib_sdma_state_s99_running);
			ss->go_s99_running = 1;
			break;
		case qib_sdma_event_e40_sw_cleaned:
			break;
		case qib_sdma_event_e50_hw_cleaned:
			break;
		case qib_sdma_event_e60_hw_halted:
			break;
		case qib_sdma_event_e70_go_idle:
			break;
		case qib_sdma_event_e7220_err_halted:
			break;
		case qib_sdma_event_e7322_err_halted:
			break;
		case qib_sdma_event_e90_timer_tick:
			break;
		}
		break;

	case qib_sdma_state_s30_sw_clean_up_wait:
		switch (event) {
		case qib_sdma_event_e00_go_hw_down:
			sdma_set_state(ppd, qib_sdma_state_s00_hw_down);
			break;
		case qib_sdma_event_e10_go_hw_start:
			break;
		case qib_sdma_event_e20_hw_started:
			break;
		case qib_sdma_event_e30_go_running:
			ss->go_s99_running = 1;
			break;
		case qib_sdma_event_e40_sw_cleaned:
			sdma_set_state(ppd,
				       qib_sdma_state_s10_hw_start_up_wait);
			sdma_hw_start_up(ppd);
			break;
		case qib_sdma_event_e50_hw_cleaned:
			break;
		case qib_sdma_event_e60_hw_halted:
			break;
		case qib_sdma_event_e70_go_idle:
			ss->go_s99_running = 0;
			break;
		case qib_sdma_event_e7220_err_halted:
			break;
		case qib_sdma_event_e7322_err_halted:
			break;
		case qib_sdma_event_e90_timer_tick:
			break;
		}
		break;

	case qib_sdma_state_s40_hw_clean_up_wait:
		switch (event) {
		case qib_sdma_event_e00_go_hw_down:
			sdma_set_state(ppd, qib_sdma_state_s00_hw_down);
			sdma_start_sw_clean_up(ppd);
			break;
		case qib_sdma_event_e10_go_hw_start:
			break;
		case qib_sdma_event_e20_hw_started:
			break;
		case qib_sdma_event_e30_go_running:
			ss->go_s99_running = 1;
			break;
		case qib_sdma_event_e40_sw_cleaned:
			break;
		case qib_sdma_event_e50_hw_cleaned:
			sdma_set_state(ppd,
				       qib_sdma_state_s30_sw_clean_up_wait);
			sdma_start_sw_clean_up(ppd);
			break;
		case qib_sdma_event_e60_hw_halted:
			break;
		case qib_sdma_event_e70_go_idle:
			ss->go_s99_running = 0;
			break;
		case qib_sdma_event_e7220_err_halted:
			break;
		case qib_sdma_event_e7322_err_halted:
			break;
		case qib_sdma_event_e90_timer_tick:
			break;
		}
		break;

	case qib_sdma_state_s50_hw_halt_wait:
		switch (event) {
		case qib_sdma_event_e00_go_hw_down:
			sdma_set_state(ppd, qib_sdma_state_s00_hw_down);
			sdma_start_sw_clean_up(ppd);
			break;
		case qib_sdma_event_e10_go_hw_start:
			break;
		case qib_sdma_event_e20_hw_started:
			break;
		case qib_sdma_event_e30_go_running:
			ss->go_s99_running = 1;
			break;
		case qib_sdma_event_e40_sw_cleaned:
			break;
		case qib_sdma_event_e50_hw_cleaned:
			break;
		case qib_sdma_event_e60_hw_halted:
			sdma_set_state(ppd,
				       qib_sdma_state_s40_hw_clean_up_wait);
			ppd->dd->f_sdma_hw_clean_up(ppd);
			break;
		case qib_sdma_event_e70_go_idle:
			ss->go_s99_running = 0;
			break;
		case qib_sdma_event_e7220_err_halted:
			break;
		case qib_sdma_event_e7322_err_halted:
			break;
		case qib_sdma_event_e90_timer_tick:
			break;
		}
		break;

	case qib_sdma_state_s99_running:
		switch (event) {
		case qib_sdma_event_e00_go_hw_down:
			sdma_set_state(ppd, qib_sdma_state_s00_hw_down);
			sdma_start_sw_clean_up(ppd);
			break;
		case qib_sdma_event_e10_go_hw_start:
			break;
		case qib_sdma_event_e20_hw_started:
			break;
		case qib_sdma_event_e30_go_running:
			break;
		case qib_sdma_event_e40_sw_cleaned:
			break;
		case qib_sdma_event_e50_hw_cleaned:
			break;
		case qib_sdma_event_e60_hw_halted:
			sdma_set_state(ppd,
				       qib_sdma_state_s30_sw_clean_up_wait);
			sdma_start_sw_clean_up(ppd);
			break;
		case qib_sdma_event_e70_go_idle:
			sdma_set_state(ppd, qib_sdma_state_s50_hw_halt_wait);
			ss->go_s99_running = 0;
			break;
		case qib_sdma_event_e7220_err_halted:
			sdma_set_state(ppd,
				       qib_sdma_state_s30_sw_clean_up_wait);
			sdma_start_sw_clean_up(ppd);
			break;
		case qib_sdma_event_e7322_err_halted:
			sdma_set_state(ppd, qib_sdma_state_s50_hw_halt_wait);
			break;
		case qib_sdma_event_e90_timer_tick:
			break;
		}
		break;
	}

	ss->last_event = event;
}