summaryrefslogtreecommitdiff
path: root/drivers/infiniband/hw/hfi1/user_sdma.c
blob: 6a4aa59c0cbccabd91e49fc20990c875dfb85626 (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
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
// SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause
/*
 * Copyright(c) 2020 - 2023 Cornelis Networks, Inc.
 * Copyright(c) 2015 - 2018 Intel Corporation.
 */

#include <linux/mm.h>
#include <linux/types.h>
#include <linux/device.h>
#include <linux/dmapool.h>
#include <linux/slab.h>
#include <linux/list.h>
#include <linux/highmem.h>
#include <linux/io.h>
#include <linux/uio.h>
#include <linux/rbtree.h>
#include <linux/spinlock.h>
#include <linux/delay.h>
#include <linux/kthread.h>
#include <linux/mmu_context.h>
#include <linux/module.h>
#include <linux/vmalloc.h>
#include <linux/string.h>

#include "hfi.h"
#include "sdma.h"
#include "user_sdma.h"
#include "verbs.h"  /* for the headers */
#include "common.h" /* for struct hfi1_tid_info */
#include "trace.h"

static uint hfi1_sdma_comp_ring_size = 128;
module_param_named(sdma_comp_size, hfi1_sdma_comp_ring_size, uint, S_IRUGO);
MODULE_PARM_DESC(sdma_comp_size, "Size of User SDMA completion ring. Default: 128");

static unsigned initial_pkt_count = 8;

static int user_sdma_send_pkts(struct user_sdma_request *req, u16 maxpkts);
static void user_sdma_txreq_cb(struct sdma_txreq *txreq, int status);
static inline void pq_update(struct hfi1_user_sdma_pkt_q *pq);
static void user_sdma_free_request(struct user_sdma_request *req);
static int check_header_template(struct user_sdma_request *req,
				 struct hfi1_pkt_header *hdr, u32 lrhlen,
				 u32 datalen);
static int set_txreq_header(struct user_sdma_request *req,
			    struct user_sdma_txreq *tx, u32 datalen);
static int set_txreq_header_ahg(struct user_sdma_request *req,
				struct user_sdma_txreq *tx, u32 len);
static inline void set_comp_state(struct hfi1_user_sdma_pkt_q *pq,
				  struct hfi1_user_sdma_comp_q *cq,
				  u16 idx, enum hfi1_sdma_comp_state state,
				  int ret);
static inline u32 set_pkt_bth_psn(__be32 bthpsn, u8 expct, u32 frags);
static inline u32 get_lrh_len(struct hfi1_pkt_header, u32 len);

static int defer_packet_queue(
	struct sdma_engine *sde,
	struct iowait_work *wait,
	struct sdma_txreq *txreq,
	uint seq,
	bool pkts_sent);
static void activate_packet_queue(struct iowait *wait, int reason);

static int defer_packet_queue(
	struct sdma_engine *sde,
	struct iowait_work *wait,
	struct sdma_txreq *txreq,
	uint seq,
	bool pkts_sent)
{
	struct hfi1_user_sdma_pkt_q *pq =
		container_of(wait->iow, struct hfi1_user_sdma_pkt_q, busy);

	write_seqlock(&sde->waitlock);
	trace_hfi1_usdma_defer(pq, sde, &pq->busy);
	if (sdma_progress(sde, seq, txreq))
		goto eagain;
	/*
	 * We are assuming that if the list is enqueued somewhere, it
	 * is to the dmawait list since that is the only place where
	 * it is supposed to be enqueued.
	 */
	xchg(&pq->state, SDMA_PKT_Q_DEFERRED);
	if (list_empty(&pq->busy.list)) {
		pq->busy.lock = &sde->waitlock;
		iowait_get_priority(&pq->busy);
		iowait_queue(pkts_sent, &pq->busy, &sde->dmawait);
	}
	write_sequnlock(&sde->waitlock);
	return -EBUSY;
eagain:
	write_sequnlock(&sde->waitlock);
	return -EAGAIN;
}

static void activate_packet_queue(struct iowait *wait, int reason)
{
	struct hfi1_user_sdma_pkt_q *pq =
		container_of(wait, struct hfi1_user_sdma_pkt_q, busy);

	trace_hfi1_usdma_activate(pq, wait, reason);
	xchg(&pq->state, SDMA_PKT_Q_ACTIVE);
	wake_up(&wait->wait_dma);
};

int hfi1_user_sdma_alloc_queues(struct hfi1_ctxtdata *uctxt,
				struct hfi1_filedata *fd)
{
	int ret = -ENOMEM;
	char buf[64];
	struct hfi1_devdata *dd;
	struct hfi1_user_sdma_comp_q *cq;
	struct hfi1_user_sdma_pkt_q *pq;

	if (!uctxt || !fd)
		return -EBADF;

	if (!hfi1_sdma_comp_ring_size)
		return -EINVAL;

	dd = uctxt->dd;

	pq = kzalloc(sizeof(*pq), GFP_KERNEL);
	if (!pq)
		return -ENOMEM;
	pq->dd = dd;
	pq->ctxt = uctxt->ctxt;
	pq->subctxt = fd->subctxt;
	pq->n_max_reqs = hfi1_sdma_comp_ring_size;
	atomic_set(&pq->n_reqs, 0);
	init_waitqueue_head(&pq->wait);
	atomic_set(&pq->n_locked, 0);

	iowait_init(&pq->busy, 0, NULL, NULL, defer_packet_queue,
		    activate_packet_queue, NULL, NULL);
	pq->reqidx = 0;

	pq->reqs = kcalloc(hfi1_sdma_comp_ring_size,
			   sizeof(*pq->reqs),
			   GFP_KERNEL);
	if (!pq->reqs)
		goto pq_reqs_nomem;

	pq->req_in_use = bitmap_zalloc(hfi1_sdma_comp_ring_size, GFP_KERNEL);
	if (!pq->req_in_use)
		goto pq_reqs_no_in_use;

	snprintf(buf, 64, "txreq-kmem-cache-%u-%u-%u", dd->unit, uctxt->ctxt,
		 fd->subctxt);
	pq->txreq_cache = kmem_cache_create(buf,
					    sizeof(struct user_sdma_txreq),
					    L1_CACHE_BYTES,
					    SLAB_HWCACHE_ALIGN,
					    NULL);
	if (!pq->txreq_cache) {
		dd_dev_err(dd, "[%u] Failed to allocate TxReq cache\n",
			   uctxt->ctxt);
		goto pq_txreq_nomem;
	}

	cq = kzalloc(sizeof(*cq), GFP_KERNEL);
	if (!cq)
		goto cq_nomem;

	cq->comps = vmalloc_user(PAGE_ALIGN(sizeof(*cq->comps)
				 * hfi1_sdma_comp_ring_size));
	if (!cq->comps)
		goto cq_comps_nomem;

	cq->nentries = hfi1_sdma_comp_ring_size;

	ret = hfi1_init_system_pinning(pq);
	if (ret)
		goto pq_mmu_fail;

	rcu_assign_pointer(fd->pq, pq);
	fd->cq = cq;

	return 0;

pq_mmu_fail:
	vfree(cq->comps);
cq_comps_nomem:
	kfree(cq);
cq_nomem:
	kmem_cache_destroy(pq->txreq_cache);
pq_txreq_nomem:
	bitmap_free(pq->req_in_use);
pq_reqs_no_in_use:
	kfree(pq->reqs);
pq_reqs_nomem:
	kfree(pq);

	return ret;
}

static void flush_pq_iowait(struct hfi1_user_sdma_pkt_q *pq)
{
	unsigned long flags;
	seqlock_t *lock = pq->busy.lock;

	if (!lock)
		return;
	write_seqlock_irqsave(lock, flags);
	if (!list_empty(&pq->busy.list)) {
		list_del_init(&pq->busy.list);
		pq->busy.lock = NULL;
	}
	write_sequnlock_irqrestore(lock, flags);
}

int hfi1_user_sdma_free_queues(struct hfi1_filedata *fd,
			       struct hfi1_ctxtdata *uctxt)
{
	struct hfi1_user_sdma_pkt_q *pq;

	trace_hfi1_sdma_user_free_queues(uctxt->dd, uctxt->ctxt, fd->subctxt);

	spin_lock(&fd->pq_rcu_lock);
	pq = srcu_dereference_check(fd->pq, &fd->pq_srcu,
				    lockdep_is_held(&fd->pq_rcu_lock));
	if (pq) {
		rcu_assign_pointer(fd->pq, NULL);
		spin_unlock(&fd->pq_rcu_lock);
		synchronize_srcu(&fd->pq_srcu);
		/* at this point there can be no more new requests */
		iowait_sdma_drain(&pq->busy);
		/* Wait until all requests have been freed. */
		wait_event_interruptible(
			pq->wait,
			!atomic_read(&pq->n_reqs));
		kfree(pq->reqs);
		hfi1_free_system_pinning(pq);
		bitmap_free(pq->req_in_use);
		kmem_cache_destroy(pq->txreq_cache);
		flush_pq_iowait(pq);
		kfree(pq);
	} else {
		spin_unlock(&fd->pq_rcu_lock);
	}
	if (fd->cq) {
		vfree(fd->cq->comps);
		kfree(fd->cq);
		fd->cq = NULL;
	}
	return 0;
}

static u8 dlid_to_selector(u16 dlid)
{
	static u8 mapping[256];
	static int initialized;
	static u8 next;
	int hash;

	if (!initialized) {
		memset(mapping, 0xFF, 256);
		initialized = 1;
	}

	hash = ((dlid >> 8) ^ dlid) & 0xFF;
	if (mapping[hash] == 0xFF) {
		mapping[hash] = next;
		next = (next + 1) & 0x7F;
	}

	return mapping[hash];
}

/**
 * hfi1_user_sdma_process_request() - Process and start a user sdma request
 * @fd: valid file descriptor
 * @iovec: array of io vectors to process
 * @dim: overall iovec array size
 * @count: number of io vector array entries processed
 */
int hfi1_user_sdma_process_request(struct hfi1_filedata *fd,
				   struct iovec *iovec, unsigned long dim,
				   unsigned long *count)
{
	int ret = 0, i;
	struct hfi1_ctxtdata *uctxt = fd->uctxt;
	struct hfi1_user_sdma_pkt_q *pq =
		srcu_dereference(fd->pq, &fd->pq_srcu);
	struct hfi1_user_sdma_comp_q *cq = fd->cq;
	struct hfi1_devdata *dd = pq->dd;
	unsigned long idx = 0;
	u8 pcount = initial_pkt_count;
	struct sdma_req_info info;
	struct user_sdma_request *req;
	u8 opcode, sc, vl;
	u16 pkey;
	u32 slid;
	u16 dlid;
	u32 selector;

	if (iovec[idx].iov_len < sizeof(info) + sizeof(req->hdr)) {
		hfi1_cdbg(
		   SDMA,
		   "[%u:%u:%u] First vector not big enough for header %lu/%lu",
		   dd->unit, uctxt->ctxt, fd->subctxt,
		   iovec[idx].iov_len, sizeof(info) + sizeof(req->hdr));
		return -EINVAL;
	}
	ret = copy_from_user(&info, iovec[idx].iov_base, sizeof(info));
	if (ret) {
		hfi1_cdbg(SDMA, "[%u:%u:%u] Failed to copy info QW (%d)",
			  dd->unit, uctxt->ctxt, fd->subctxt, ret);
		return -EFAULT;
	}

	trace_hfi1_sdma_user_reqinfo(dd, uctxt->ctxt, fd->subctxt,
				     (u16 *)&info);
	if (info.comp_idx >= hfi1_sdma_comp_ring_size) {
		hfi1_cdbg(SDMA,
			  "[%u:%u:%u:%u] Invalid comp index",
			  dd->unit, uctxt->ctxt, fd->subctxt, info.comp_idx);
		return -EINVAL;
	}

	/*
	 * Sanity check the header io vector count.  Need at least 1 vector
	 * (header) and cannot be larger than the actual io vector count.
	 */
	if (req_iovcnt(info.ctrl) < 1 || req_iovcnt(info.ctrl) > dim) {
		hfi1_cdbg(SDMA,
			  "[%u:%u:%u:%u] Invalid iov count %d, dim %ld",
			  dd->unit, uctxt->ctxt, fd->subctxt, info.comp_idx,
			  req_iovcnt(info.ctrl), dim);
		return -EINVAL;
	}

	if (!info.fragsize) {
		hfi1_cdbg(SDMA,
			  "[%u:%u:%u:%u] Request does not specify fragsize",
			  dd->unit, uctxt->ctxt, fd->subctxt, info.comp_idx);
		return -EINVAL;
	}

	/* Try to claim the request. */
	if (test_and_set_bit(info.comp_idx, pq->req_in_use)) {
		hfi1_cdbg(SDMA, "[%u:%u:%u] Entry %u is in use",
			  dd->unit, uctxt->ctxt, fd->subctxt,
			  info.comp_idx);
		return -EBADSLT;
	}
	/*
	 * All safety checks have been done and this request has been claimed.
	 */
	trace_hfi1_sdma_user_process_request(dd, uctxt->ctxt, fd->subctxt,
					     info.comp_idx);
	req = pq->reqs + info.comp_idx;
	req->data_iovs = req_iovcnt(info.ctrl) - 1; /* subtract header vector */
	req->data_len  = 0;
	req->pq = pq;
	req->cq = cq;
	req->ahg_idx = -1;
	req->iov_idx = 0;
	req->sent = 0;
	req->seqnum = 0;
	req->seqcomp = 0;
	req->seqsubmitted = 0;
	req->tids = NULL;
	req->has_error = 0;
	INIT_LIST_HEAD(&req->txps);

	memcpy(&req->info, &info, sizeof(info));

	/* The request is initialized, count it */
	atomic_inc(&pq->n_reqs);

	if (req_opcode(info.ctrl) == EXPECTED) {
		/* expected must have a TID info and at least one data vector */
		if (req->data_iovs < 2) {
			SDMA_DBG(req,
				 "Not enough vectors for expected request");
			ret = -EINVAL;
			goto free_req;
		}
		req->data_iovs--;
	}

	if (!info.npkts || req->data_iovs > MAX_VECTORS_PER_REQ) {
		SDMA_DBG(req, "Too many vectors (%u/%u)", req->data_iovs,
			 MAX_VECTORS_PER_REQ);
		ret = -EINVAL;
		goto free_req;
	}

	/* Copy the header from the user buffer */
	ret = copy_from_user(&req->hdr, iovec[idx].iov_base + sizeof(info),
			     sizeof(req->hdr));
	if (ret) {
		SDMA_DBG(req, "Failed to copy header template (%d)", ret);
		ret = -EFAULT;
		goto free_req;
	}

	/* If Static rate control is not enabled, sanitize the header. */
	if (!HFI1_CAP_IS_USET(STATIC_RATE_CTRL))
		req->hdr.pbc[2] = 0;

	/* Validate the opcode. Do not trust packets from user space blindly. */
	opcode = (be32_to_cpu(req->hdr.bth[0]) >> 24) & 0xff;
	if ((opcode & USER_OPCODE_CHECK_MASK) !=
	     USER_OPCODE_CHECK_VAL) {
		SDMA_DBG(req, "Invalid opcode (%d)", opcode);
		ret = -EINVAL;
		goto free_req;
	}
	/*
	 * Validate the vl. Do not trust packets from user space blindly.
	 * VL comes from PBC, SC comes from LRH, and the VL needs to
	 * match the SC look up.
	 */
	vl = (le16_to_cpu(req->hdr.pbc[0]) >> 12) & 0xF;
	sc = (((be16_to_cpu(req->hdr.lrh[0]) >> 12) & 0xF) |
	      (((le16_to_cpu(req->hdr.pbc[1]) >> 14) & 0x1) << 4));
	if (vl >= dd->pport->vls_operational ||
	    vl != sc_to_vlt(dd, sc)) {
		SDMA_DBG(req, "Invalid SC(%u)/VL(%u)", sc, vl);
		ret = -EINVAL;
		goto free_req;
	}

	/* Checking P_KEY for requests from user-space */
	pkey = (u16)be32_to_cpu(req->hdr.bth[0]);
	slid = be16_to_cpu(req->hdr.lrh[3]);
	if (egress_pkey_check(dd->pport, slid, pkey, sc, PKEY_CHECK_INVALID)) {
		ret = -EINVAL;
		goto free_req;
	}

	/*
	 * Also should check the BTH.lnh. If it says the next header is GRH then
	 * the RXE parsing will be off and will land in the middle of the KDETH
	 * or miss it entirely.
	 */
	if ((be16_to_cpu(req->hdr.lrh[0]) & 0x3) == HFI1_LRH_GRH) {
		SDMA_DBG(req, "User tried to pass in a GRH");
		ret = -EINVAL;
		goto free_req;
	}

	req->koffset = le32_to_cpu(req->hdr.kdeth.swdata[6]);
	/*
	 * Calculate the initial TID offset based on the values of
	 * KDETH.OFFSET and KDETH.OM that are passed in.
	 */
	req->tidoffset = KDETH_GET(req->hdr.kdeth.ver_tid_offset, OFFSET) *
		(KDETH_GET(req->hdr.kdeth.ver_tid_offset, OM) ?
		 KDETH_OM_LARGE : KDETH_OM_SMALL);
	trace_hfi1_sdma_user_initial_tidoffset(dd, uctxt->ctxt, fd->subctxt,
					       info.comp_idx, req->tidoffset);
	idx++;

	/* Save all the IO vector structures */
	for (i = 0; i < req->data_iovs; i++) {
		req->iovs[i].offset = 0;
		INIT_LIST_HEAD(&req->iovs[i].list);
		memcpy(&req->iovs[i].iov,
		       iovec + idx++,
		       sizeof(req->iovs[i].iov));
		if (req->iovs[i].iov.iov_len == 0) {
			ret = -EINVAL;
			goto free_req;
		}
		req->data_len += req->iovs[i].iov.iov_len;
	}
	trace_hfi1_sdma_user_data_length(dd, uctxt->ctxt, fd->subctxt,
					 info.comp_idx, req->data_len);
	if (pcount > req->info.npkts)
		pcount = req->info.npkts;
	/*
	 * Copy any TID info
	 * User space will provide the TID info only when the
	 * request type is EXPECTED. This is true even if there is
	 * only one packet in the request and the header is already
	 * setup. The reason for the singular TID case is that the
	 * driver needs to perform safety checks.
	 */
	if (req_opcode(req->info.ctrl) == EXPECTED) {
		u16 ntids = iovec[idx].iov_len / sizeof(*req->tids);
		u32 *tmp;

		if (!ntids || ntids > MAX_TID_PAIR_ENTRIES) {
			ret = -EINVAL;
			goto free_req;
		}

		/*
		 * We have to copy all of the tids because they may vary
		 * in size and, therefore, the TID count might not be
		 * equal to the pkt count. However, there is no way to
		 * tell at this point.
		 */
		tmp = memdup_user(iovec[idx].iov_base,
				  ntids * sizeof(*req->tids));
		if (IS_ERR(tmp)) {
			ret = PTR_ERR(tmp);
			SDMA_DBG(req, "Failed to copy %d TIDs (%d)",
				 ntids, ret);
			goto free_req;
		}
		req->tids = tmp;
		req->n_tids = ntids;
		req->tididx = 0;
		idx++;
	}

	dlid = be16_to_cpu(req->hdr.lrh[1]);
	selector = dlid_to_selector(dlid);
	selector += uctxt->ctxt + fd->subctxt;
	req->sde = sdma_select_user_engine(dd, selector, vl);

	if (!req->sde || !sdma_running(req->sde)) {
		ret = -ECOMM;
		goto free_req;
	}

	/* We don't need an AHG entry if the request contains only one packet */
	if (req->info.npkts > 1 && HFI1_CAP_IS_USET(SDMA_AHG))
		req->ahg_idx = sdma_ahg_alloc(req->sde);

	set_comp_state(pq, cq, info.comp_idx, QUEUED, 0);
	pq->state = SDMA_PKT_Q_ACTIVE;

	/*
	 * This is a somewhat blocking send implementation.
	 * The driver will block the caller until all packets of the
	 * request have been submitted to the SDMA engine. However, it
	 * will not wait for send completions.
	 */
	while (req->seqsubmitted != req->info.npkts) {
		ret = user_sdma_send_pkts(req, pcount);
		if (ret < 0) {
			int we_ret;

			if (ret != -EBUSY)
				goto free_req;
			we_ret = wait_event_interruptible_timeout(
				pq->busy.wait_dma,
				pq->state == SDMA_PKT_Q_ACTIVE,
				msecs_to_jiffies(
					SDMA_IOWAIT_TIMEOUT));
			trace_hfi1_usdma_we(pq, we_ret);
			if (we_ret <= 0)
				flush_pq_iowait(pq);
		}
	}
	*count += idx;
	return 0;
free_req:
	/*
	 * If the submitted seqsubmitted == npkts, the completion routine
	 * controls the final state.  If sequbmitted < npkts, wait for any
	 * outstanding packets to finish before cleaning up.
	 */
	if (req->seqsubmitted < req->info.npkts) {
		if (req->seqsubmitted)
			wait_event(pq->busy.wait_dma,
				   (req->seqcomp == req->seqsubmitted - 1));
		user_sdma_free_request(req);
		pq_update(pq);
		set_comp_state(pq, cq, info.comp_idx, ERROR, ret);
	}
	return ret;
}

static inline u32 compute_data_length(struct user_sdma_request *req,
				      struct user_sdma_txreq *tx)
{
	/*
	 * Determine the proper size of the packet data.
	 * The size of the data of the first packet is in the header
	 * template. However, it includes the header and ICRC, which need
	 * to be subtracted.
	 * The minimum representable packet data length in a header is 4 bytes,
	 * therefore, when the data length request is less than 4 bytes, there's
	 * only one packet, and the packet data length is equal to that of the
	 * request data length.
	 * The size of the remaining packets is the minimum of the frag
	 * size (MTU) or remaining data in the request.
	 */
	u32 len;

	if (!req->seqnum) {
		if (req->data_len < sizeof(u32))
			len = req->data_len;
		else
			len = ((be16_to_cpu(req->hdr.lrh[2]) << 2) -
			       (sizeof(tx->hdr) - 4));
	} else if (req_opcode(req->info.ctrl) == EXPECTED) {
		u32 tidlen = EXP_TID_GET(req->tids[req->tididx], LEN) *
			PAGE_SIZE;
		/*
		 * Get the data length based on the remaining space in the
		 * TID pair.
		 */
		len = min(tidlen - req->tidoffset, (u32)req->info.fragsize);
		/* If we've filled up the TID pair, move to the next one. */
		if (unlikely(!len) && ++req->tididx < req->n_tids &&
		    req->tids[req->tididx]) {
			tidlen = EXP_TID_GET(req->tids[req->tididx],
					     LEN) * PAGE_SIZE;
			req->tidoffset = 0;
			len = min_t(u32, tidlen, req->info.fragsize);
		}
		/*
		 * Since the TID pairs map entire pages, make sure that we
		 * are not going to try to send more data that we have
		 * remaining.
		 */
		len = min(len, req->data_len - req->sent);
	} else {
		len = min(req->data_len - req->sent, (u32)req->info.fragsize);
	}
	trace_hfi1_sdma_user_compute_length(req->pq->dd,
					    req->pq->ctxt,
					    req->pq->subctxt,
					    req->info.comp_idx,
					    len);
	return len;
}

static inline u32 pad_len(u32 len)
{
	if (len & (sizeof(u32) - 1))
		len += sizeof(u32) - (len & (sizeof(u32) - 1));
	return len;
}

static inline u32 get_lrh_len(struct hfi1_pkt_header hdr, u32 len)
{
	/* (Size of complete header - size of PBC) + 4B ICRC + data length */
	return ((sizeof(hdr) - sizeof(hdr.pbc)) + 4 + len);
}

static int user_sdma_txadd_ahg(struct user_sdma_request *req,
			       struct user_sdma_txreq *tx,
			       u32 datalen)
{
	int ret;
	u16 pbclen = le16_to_cpu(req->hdr.pbc[0]);
	u32 lrhlen = get_lrh_len(req->hdr, pad_len(datalen));
	struct hfi1_user_sdma_pkt_q *pq = req->pq;

	/*
	 * Copy the request header into the tx header
	 * because the HW needs a cacheline-aligned
	 * address.
	 * This copy can be optimized out if the hdr
	 * member of user_sdma_request were also
	 * cacheline aligned.
	 */
	memcpy(&tx->hdr, &req->hdr, sizeof(tx->hdr));
	if (PBC2LRH(pbclen) != lrhlen) {
		pbclen = (pbclen & 0xf000) | LRH2PBC(lrhlen);
		tx->hdr.pbc[0] = cpu_to_le16(pbclen);
	}
	ret = check_header_template(req, &tx->hdr, lrhlen, datalen);
	if (ret)
		return ret;
	ret = sdma_txinit_ahg(&tx->txreq, SDMA_TXREQ_F_AHG_COPY,
			      sizeof(tx->hdr) + datalen, req->ahg_idx,
			      0, NULL, 0, user_sdma_txreq_cb);
	if (ret)
		return ret;
	ret = sdma_txadd_kvaddr(pq->dd, &tx->txreq, &tx->hdr, sizeof(tx->hdr));
	if (ret)
		sdma_txclean(pq->dd, &tx->txreq);
	return ret;
}

static int user_sdma_send_pkts(struct user_sdma_request *req, u16 maxpkts)
{
	int ret = 0;
	u16 count;
	unsigned npkts = 0;
	struct user_sdma_txreq *tx = NULL;
	struct hfi1_user_sdma_pkt_q *pq = NULL;
	struct user_sdma_iovec *iovec = NULL;

	if (!req->pq)
		return -EINVAL;

	pq = req->pq;

	/* If tx completion has reported an error, we are done. */
	if (READ_ONCE(req->has_error))
		return -EFAULT;

	/*
	 * Check if we might have sent the entire request already
	 */
	if (unlikely(req->seqnum == req->info.npkts)) {
		if (!list_empty(&req->txps))
			goto dosend;
		return ret;
	}

	if (!maxpkts || maxpkts > req->info.npkts - req->seqnum)
		maxpkts = req->info.npkts - req->seqnum;

	while (npkts < maxpkts) {
		u32 datalen = 0;

		/*
		 * Check whether any of the completions have come back
		 * with errors. If so, we are not going to process any
		 * more packets from this request.
		 */
		if (READ_ONCE(req->has_error))
			return -EFAULT;

		tx = kmem_cache_alloc(pq->txreq_cache, GFP_KERNEL);
		if (!tx)
			return -ENOMEM;

		tx->flags = 0;
		tx->req = req;
		INIT_LIST_HEAD(&tx->list);

		/*
		 * For the last packet set the ACK request
		 * and disable header suppression.
		 */
		if (req->seqnum == req->info.npkts - 1)
			tx->flags |= (TXREQ_FLAGS_REQ_ACK |
				      TXREQ_FLAGS_REQ_DISABLE_SH);

		/*
		 * Calculate the payload size - this is min of the fragment
		 * (MTU) size or the remaining bytes in the request but only
		 * if we have payload data.
		 */
		if (req->data_len) {
			iovec = &req->iovs[req->iov_idx];
			if (READ_ONCE(iovec->offset) == iovec->iov.iov_len) {
				if (++req->iov_idx == req->data_iovs) {
					ret = -EFAULT;
					goto free_tx;
				}
				iovec = &req->iovs[req->iov_idx];
				WARN_ON(iovec->offset);
			}

			datalen = compute_data_length(req, tx);

			/*
			 * Disable header suppression for the payload <= 8DWS.
			 * If there is an uncorrectable error in the receive
			 * data FIFO when the received payload size is less than
			 * or equal to 8DWS then the RxDmaDataFifoRdUncErr is
			 * not reported.There is set RHF.EccErr if the header
			 * is not suppressed.
			 */
			if (!datalen) {
				SDMA_DBG(req,
					 "Request has data but pkt len is 0");
				ret = -EFAULT;
				goto free_tx;
			} else if (datalen <= 32) {
				tx->flags |= TXREQ_FLAGS_REQ_DISABLE_SH;
			}
		}

		if (req->ahg_idx >= 0) {
			if (!req->seqnum) {
				ret = user_sdma_txadd_ahg(req, tx, datalen);
				if (ret)
					goto free_tx;
			} else {
				int changes;

				changes = set_txreq_header_ahg(req, tx,
							       datalen);
				if (changes < 0) {
					ret = changes;
					goto free_tx;
				}
			}
		} else {
			ret = sdma_txinit(&tx->txreq, 0, sizeof(req->hdr) +
					  datalen, user_sdma_txreq_cb);
			if (ret)
				goto free_tx;
			/*
			 * Modify the header for this packet. This only needs
			 * to be done if we are not going to use AHG. Otherwise,
			 * the HW will do it based on the changes we gave it
			 * during sdma_txinit_ahg().
			 */
			ret = set_txreq_header(req, tx, datalen);
			if (ret)
				goto free_txreq;
		}

		req->koffset += datalen;
		if (req_opcode(req->info.ctrl) == EXPECTED)
			req->tidoffset += datalen;
		req->sent += datalen;
		while (datalen) {
			ret = hfi1_add_pages_to_sdma_packet(req, tx, iovec,
							    &datalen);
			if (ret)
				goto free_txreq;
			iovec = &req->iovs[req->iov_idx];
		}
		list_add_tail(&tx->txreq.list, &req->txps);
		/*
		 * It is important to increment this here as it is used to
		 * generate the BTH.PSN and, therefore, can't be bulk-updated
		 * outside of the loop.
		 */
		tx->seqnum = req->seqnum++;
		npkts++;
	}
dosend:
	ret = sdma_send_txlist(req->sde,
			       iowait_get_ib_work(&pq->busy),
			       &req->txps, &count);
	req->seqsubmitted += count;
	if (req->seqsubmitted == req->info.npkts) {
		/*
		 * The txreq has already been submitted to the HW queue
		 * so we can free the AHG entry now. Corruption will not
		 * happen due to the sequential manner in which
		 * descriptors are processed.
		 */
		if (req->ahg_idx >= 0)
			sdma_ahg_free(req->sde, req->ahg_idx);
	}
	return ret;

free_txreq:
	sdma_txclean(pq->dd, &tx->txreq);
free_tx:
	kmem_cache_free(pq->txreq_cache, tx);
	return ret;
}

static int check_header_template(struct user_sdma_request *req,
				 struct hfi1_pkt_header *hdr, u32 lrhlen,
				 u32 datalen)
{
	/*
	 * Perform safety checks for any type of packet:
	 *    - transfer size is multiple of 64bytes
	 *    - packet length is multiple of 4 bytes
	 *    - packet length is not larger than MTU size
	 *
	 * These checks are only done for the first packet of the
	 * transfer since the header is "given" to us by user space.
	 * For the remainder of the packets we compute the values.
	 */
	if (req->info.fragsize % PIO_BLOCK_SIZE || lrhlen & 0x3 ||
	    lrhlen > get_lrh_len(*hdr, req->info.fragsize))
		return -EINVAL;

	if (req_opcode(req->info.ctrl) == EXPECTED) {
		/*
		 * The header is checked only on the first packet. Furthermore,
		 * we ensure that at least one TID entry is copied when the
		 * request is submitted. Therefore, we don't have to verify that
		 * tididx points to something sane.
		 */
		u32 tidval = req->tids[req->tididx],
			tidlen = EXP_TID_GET(tidval, LEN) * PAGE_SIZE,
			tididx = EXP_TID_GET(tidval, IDX),
			tidctrl = EXP_TID_GET(tidval, CTRL),
			tidoff;
		__le32 kval = hdr->kdeth.ver_tid_offset;

		tidoff = KDETH_GET(kval, OFFSET) *
			  (KDETH_GET(req->hdr.kdeth.ver_tid_offset, OM) ?
			   KDETH_OM_LARGE : KDETH_OM_SMALL);
		/*
		 * Expected receive packets have the following
		 * additional checks:
		 *     - offset is not larger than the TID size
		 *     - TIDCtrl values match between header and TID array
		 *     - TID indexes match between header and TID array
		 */
		if ((tidoff + datalen > tidlen) ||
		    KDETH_GET(kval, TIDCTRL) != tidctrl ||
		    KDETH_GET(kval, TID) != tididx)
			return -EINVAL;
	}
	return 0;
}

/*
 * Correctly set the BTH.PSN field based on type of
 * transfer - eager packets can just increment the PSN but
 * expected packets encode generation and sequence in the
 * BTH.PSN field so just incrementing will result in errors.
 */
static inline u32 set_pkt_bth_psn(__be32 bthpsn, u8 expct, u32 frags)
{
	u32 val = be32_to_cpu(bthpsn),
		mask = (HFI1_CAP_IS_KSET(EXTENDED_PSN) ? 0x7fffffffull :
			0xffffffull),
		psn = val & mask;
	if (expct)
		psn = (psn & ~HFI1_KDETH_BTH_SEQ_MASK) |
			((psn + frags) & HFI1_KDETH_BTH_SEQ_MASK);
	else
		psn = psn + frags;
	return psn & mask;
}

static int set_txreq_header(struct user_sdma_request *req,
			    struct user_sdma_txreq *tx, u32 datalen)
{
	struct hfi1_user_sdma_pkt_q *pq = req->pq;
	struct hfi1_pkt_header *hdr = &tx->hdr;
	u8 omfactor; /* KDETH.OM */
	u16 pbclen;
	int ret;
	u32 tidval = 0, lrhlen = get_lrh_len(*hdr, pad_len(datalen));

	/* Copy the header template to the request before modification */
	memcpy(hdr, &req->hdr, sizeof(*hdr));

	/*
	 * Check if the PBC and LRH length are mismatched. If so
	 * adjust both in the header.
	 */
	pbclen = le16_to_cpu(hdr->pbc[0]);
	if (PBC2LRH(pbclen) != lrhlen) {
		pbclen = (pbclen & 0xf000) | LRH2PBC(lrhlen);
		hdr->pbc[0] = cpu_to_le16(pbclen);
		hdr->lrh[2] = cpu_to_be16(lrhlen >> 2);
		/*
		 * Third packet
		 * This is the first packet in the sequence that has
		 * a "static" size that can be used for the rest of
		 * the packets (besides the last one).
		 */
		if (unlikely(req->seqnum == 2)) {
			/*
			 * From this point on the lengths in both the
			 * PBC and LRH are the same until the last
			 * packet.
			 * Adjust the template so we don't have to update
			 * every packet
			 */
			req->hdr.pbc[0] = hdr->pbc[0];
			req->hdr.lrh[2] = hdr->lrh[2];
		}
	}
	/*
	 * We only have to modify the header if this is not the
	 * first packet in the request. Otherwise, we use the
	 * header given to us.
	 */
	if (unlikely(!req->seqnum)) {
		ret = check_header_template(req, hdr, lrhlen, datalen);
		if (ret)
			return ret;
		goto done;
	}

	hdr->bth[2] = cpu_to_be32(
		set_pkt_bth_psn(hdr->bth[2],
				(req_opcode(req->info.ctrl) == EXPECTED),
				req->seqnum));

	/* Set ACK request on last packet */
	if (unlikely(tx->flags & TXREQ_FLAGS_REQ_ACK))
		hdr->bth[2] |= cpu_to_be32(1UL << 31);

	/* Set the new offset */
	hdr->kdeth.swdata[6] = cpu_to_le32(req->koffset);
	/* Expected packets have to fill in the new TID information */
	if (req_opcode(req->info.ctrl) == EXPECTED) {
		tidval = req->tids[req->tididx];
		/*
		 * If the offset puts us at the end of the current TID,
		 * advance everything.
		 */
		if ((req->tidoffset) == (EXP_TID_GET(tidval, LEN) *
					 PAGE_SIZE)) {
			req->tidoffset = 0;
			/*
			 * Since we don't copy all the TIDs, all at once,
			 * we have to check again.
			 */
			if (++req->tididx > req->n_tids - 1 ||
			    !req->tids[req->tididx]) {
				return -EINVAL;
			}
			tidval = req->tids[req->tididx];
		}
		omfactor = EXP_TID_GET(tidval, LEN) * PAGE_SIZE >=
			KDETH_OM_MAX_SIZE ? KDETH_OM_LARGE_SHIFT :
			KDETH_OM_SMALL_SHIFT;
		/* Set KDETH.TIDCtrl based on value for this TID. */
		KDETH_SET(hdr->kdeth.ver_tid_offset, TIDCTRL,
			  EXP_TID_GET(tidval, CTRL));
		/* Set KDETH.TID based on value for this TID */
		KDETH_SET(hdr->kdeth.ver_tid_offset, TID,
			  EXP_TID_GET(tidval, IDX));
		/* Clear KDETH.SH when DISABLE_SH flag is set */
		if (unlikely(tx->flags & TXREQ_FLAGS_REQ_DISABLE_SH))
			KDETH_SET(hdr->kdeth.ver_tid_offset, SH, 0);
		/*
		 * Set the KDETH.OFFSET and KDETH.OM based on size of
		 * transfer.
		 */
		trace_hfi1_sdma_user_tid_info(
			pq->dd, pq->ctxt, pq->subctxt, req->info.comp_idx,
			req->tidoffset, req->tidoffset >> omfactor,
			omfactor != KDETH_OM_SMALL_SHIFT);
		KDETH_SET(hdr->kdeth.ver_tid_offset, OFFSET,
			  req->tidoffset >> omfactor);
		KDETH_SET(hdr->kdeth.ver_tid_offset, OM,
			  omfactor != KDETH_OM_SMALL_SHIFT);
	}
done:
	trace_hfi1_sdma_user_header(pq->dd, pq->ctxt, pq->subctxt,
				    req->info.comp_idx, hdr, tidval);
	return sdma_txadd_kvaddr(pq->dd, &tx->txreq, hdr, sizeof(*hdr));
}

static int set_txreq_header_ahg(struct user_sdma_request *req,
				struct user_sdma_txreq *tx, u32 datalen)
{
	u32 ahg[AHG_KDETH_ARRAY_SIZE];
	int idx = 0;
	u8 omfactor; /* KDETH.OM */
	struct hfi1_user_sdma_pkt_q *pq = req->pq;
	struct hfi1_pkt_header *hdr = &req->hdr;
	u16 pbclen = le16_to_cpu(hdr->pbc[0]);
	u32 val32, tidval = 0, lrhlen = get_lrh_len(*hdr, pad_len(datalen));
	size_t array_size = ARRAY_SIZE(ahg);

	if (PBC2LRH(pbclen) != lrhlen) {
		/* PBC.PbcLengthDWs */
		idx = ahg_header_set(ahg, idx, array_size, 0, 0, 12,
				     (__force u16)cpu_to_le16(LRH2PBC(lrhlen)));
		if (idx < 0)
			return idx;
		/* LRH.PktLen (we need the full 16 bits due to byte swap) */
		idx = ahg_header_set(ahg, idx, array_size, 3, 0, 16,
				     (__force u16)cpu_to_be16(lrhlen >> 2));
		if (idx < 0)
			return idx;
	}

	/*
	 * Do the common updates
	 */
	/* BTH.PSN and BTH.A */
	val32 = (be32_to_cpu(hdr->bth[2]) + req->seqnum) &
		(HFI1_CAP_IS_KSET(EXTENDED_PSN) ? 0x7fffffff : 0xffffff);
	if (unlikely(tx->flags & TXREQ_FLAGS_REQ_ACK))
		val32 |= 1UL << 31;
	idx = ahg_header_set(ahg, idx, array_size, 6, 0, 16,
			     (__force u16)cpu_to_be16(val32 >> 16));
	if (idx < 0)
		return idx;
	idx = ahg_header_set(ahg, idx, array_size, 6, 16, 16,
			     (__force u16)cpu_to_be16(val32 & 0xffff));
	if (idx < 0)
		return idx;
	/* KDETH.Offset */
	idx = ahg_header_set(ahg, idx, array_size, 15, 0, 16,
			     (__force u16)cpu_to_le16(req->koffset & 0xffff));
	if (idx < 0)
		return idx;
	idx = ahg_header_set(ahg, idx, array_size, 15, 16, 16,
			     (__force u16)cpu_to_le16(req->koffset >> 16));
	if (idx < 0)
		return idx;
	if (req_opcode(req->info.ctrl) == EXPECTED) {
		__le16 val;

		tidval = req->tids[req->tididx];

		/*
		 * If the offset puts us at the end of the current TID,
		 * advance everything.
		 */
		if ((req->tidoffset) == (EXP_TID_GET(tidval, LEN) *
					 PAGE_SIZE)) {
			req->tidoffset = 0;
			/*
			 * Since we don't copy all the TIDs, all at once,
			 * we have to check again.
			 */
			if (++req->tididx > req->n_tids - 1 ||
			    !req->tids[req->tididx])
				return -EINVAL;
			tidval = req->tids[req->tididx];
		}
		omfactor = ((EXP_TID_GET(tidval, LEN) *
				  PAGE_SIZE) >=
				 KDETH_OM_MAX_SIZE) ? KDETH_OM_LARGE_SHIFT :
				 KDETH_OM_SMALL_SHIFT;
		/* KDETH.OM and KDETH.OFFSET (TID) */
		idx = ahg_header_set(
				ahg, idx, array_size, 7, 0, 16,
				((!!(omfactor - KDETH_OM_SMALL_SHIFT)) << 15 |
				((req->tidoffset >> omfactor)
				& 0x7fff)));
		if (idx < 0)
			return idx;
		/* KDETH.TIDCtrl, KDETH.TID, KDETH.Intr, KDETH.SH */
		val = cpu_to_le16(((EXP_TID_GET(tidval, CTRL) & 0x3) << 10) |
				   (EXP_TID_GET(tidval, IDX) & 0x3ff));

		if (unlikely(tx->flags & TXREQ_FLAGS_REQ_DISABLE_SH)) {
			val |= cpu_to_le16((KDETH_GET(hdr->kdeth.ver_tid_offset,
						      INTR) <<
					    AHG_KDETH_INTR_SHIFT));
		} else {
			val |= KDETH_GET(hdr->kdeth.ver_tid_offset, SH) ?
			       cpu_to_le16(0x1 << AHG_KDETH_SH_SHIFT) :
			       cpu_to_le16((KDETH_GET(hdr->kdeth.ver_tid_offset,
						      INTR) <<
					     AHG_KDETH_INTR_SHIFT));
		}

		idx = ahg_header_set(ahg, idx, array_size,
				     7, 16, 14, (__force u16)val);
		if (idx < 0)
			return idx;
	}

	trace_hfi1_sdma_user_header_ahg(pq->dd, pq->ctxt, pq->subctxt,
					req->info.comp_idx, req->sde->this_idx,
					req->ahg_idx, ahg, idx, tidval);
	sdma_txinit_ahg(&tx->txreq,
			SDMA_TXREQ_F_USE_AHG,
			datalen, req->ahg_idx, idx,
			ahg, sizeof(req->hdr),
			user_sdma_txreq_cb);

	return idx;
}

/**
 * user_sdma_txreq_cb() - SDMA tx request completion callback.
 * @txreq: valid sdma tx request
 * @status: success/failure of request
 *
 * Called when the SDMA progress state machine gets notification that
 * the SDMA descriptors for this tx request have been processed by the
 * DMA engine. Called in interrupt context.
 * Only do work on completed sequences.
 */
static void user_sdma_txreq_cb(struct sdma_txreq *txreq, int status)
{
	struct user_sdma_txreq *tx =
		container_of(txreq, struct user_sdma_txreq, txreq);
	struct user_sdma_request *req;
	struct hfi1_user_sdma_pkt_q *pq;
	struct hfi1_user_sdma_comp_q *cq;
	enum hfi1_sdma_comp_state state = COMPLETE;

	if (!tx->req)
		return;

	req = tx->req;
	pq = req->pq;
	cq = req->cq;

	if (status != SDMA_TXREQ_S_OK) {
		SDMA_DBG(req, "SDMA completion with error %d",
			 status);
		WRITE_ONCE(req->has_error, 1);
		state = ERROR;
	}

	req->seqcomp = tx->seqnum;
	kmem_cache_free(pq->txreq_cache, tx);

	/* sequence isn't complete?  We are done */
	if (req->seqcomp != req->info.npkts - 1)
		return;

	user_sdma_free_request(req);
	set_comp_state(pq, cq, req->info.comp_idx, state, status);
	pq_update(pq);
}

static inline void pq_update(struct hfi1_user_sdma_pkt_q *pq)
{
	if (atomic_dec_and_test(&pq->n_reqs))
		wake_up(&pq->wait);
}

static void user_sdma_free_request(struct user_sdma_request *req)
{
	if (!list_empty(&req->txps)) {
		struct sdma_txreq *t, *p;

		list_for_each_entry_safe(t, p, &req->txps, list) {
			struct user_sdma_txreq *tx =
				container_of(t, struct user_sdma_txreq, txreq);
			list_del_init(&t->list);
			sdma_txclean(req->pq->dd, t);
			kmem_cache_free(req->pq->txreq_cache, tx);
		}
	}

	kfree(req->tids);
	clear_bit(req->info.comp_idx, req->pq->req_in_use);
}

static inline void set_comp_state(struct hfi1_user_sdma_pkt_q *pq,
				  struct hfi1_user_sdma_comp_q *cq,
				  u16 idx, enum hfi1_sdma_comp_state state,
				  int ret)
{
	if (state == ERROR)
		cq->comps[idx].errcode = -ret;
	smp_wmb(); /* make sure errcode is visible first */
	cq->comps[idx].status = state;
	trace_hfi1_sdma_user_completion(pq->dd, pq->ctxt, pq->subctxt,
					idx, state, ret);
}