summaryrefslogtreecommitdiff
path: root/drivers/infiniband/sw/siw/siw_qp.c
blob: ddb2e66f9f1334fa916ea8d8de2f3da4bb680599 (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
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
// SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause

/* Authors: Bernard Metzler <bmt@zurich.ibm.com> */
/* Copyright (c) 2008-2019, IBM Corporation */

#include <linux/errno.h>
#include <linux/types.h>
#include <linux/net.h>
#include <linux/scatterlist.h>
#include <linux/llist.h>
#include <asm/barrier.h>
#include <net/tcp.h>

#include "siw.h"
#include "siw_verbs.h"
#include "siw_mem.h"

static char siw_qp_state_to_string[SIW_QP_STATE_COUNT][sizeof "TERMINATE"] = {
	[SIW_QP_STATE_IDLE] = "IDLE",
	[SIW_QP_STATE_RTR] = "RTR",
	[SIW_QP_STATE_RTS] = "RTS",
	[SIW_QP_STATE_CLOSING] = "CLOSING",
	[SIW_QP_STATE_TERMINATE] = "TERMINATE",
	[SIW_QP_STATE_ERROR] = "ERROR"
};

/*
 * iWARP (RDMAP, DDP and MPA) parameters as well as Softiwarp settings on a
 * per-RDMAP message basis. Please keep order of initializer. All MPA len
 * is initialized to minimum packet size.
 */
struct iwarp_msg_info iwarp_pktinfo[RDMAP_TERMINATE + 1] = {
	{ /* RDMAP_RDMA_WRITE */
	  .hdr_len = sizeof(struct iwarp_rdma_write),
	  .ctrl.mpa_len = htons(sizeof(struct iwarp_rdma_write) - 2),
	  .ctrl.ddp_rdmap_ctrl = DDP_FLAG_TAGGED | DDP_FLAG_LAST |
				 cpu_to_be16(DDP_VERSION << 8) |
				 cpu_to_be16(RDMAP_VERSION << 6) |
				 cpu_to_be16(RDMAP_RDMA_WRITE),
	  .rx_data = siw_proc_write },
	{ /* RDMAP_RDMA_READ_REQ */
	  .hdr_len = sizeof(struct iwarp_rdma_rreq),
	  .ctrl.mpa_len = htons(sizeof(struct iwarp_rdma_rreq) - 2),
	  .ctrl.ddp_rdmap_ctrl = DDP_FLAG_LAST | cpu_to_be16(DDP_VERSION << 8) |
				 cpu_to_be16(RDMAP_VERSION << 6) |
				 cpu_to_be16(RDMAP_RDMA_READ_REQ),
	  .rx_data = siw_proc_rreq },
	{ /* RDMAP_RDMA_READ_RESP */
	  .hdr_len = sizeof(struct iwarp_rdma_rresp),
	  .ctrl.mpa_len = htons(sizeof(struct iwarp_rdma_rresp) - 2),
	  .ctrl.ddp_rdmap_ctrl = DDP_FLAG_TAGGED | DDP_FLAG_LAST |
				 cpu_to_be16(DDP_VERSION << 8) |
				 cpu_to_be16(RDMAP_VERSION << 6) |
				 cpu_to_be16(RDMAP_RDMA_READ_RESP),
	  .rx_data = siw_proc_rresp },
	{ /* RDMAP_SEND */
	  .hdr_len = sizeof(struct iwarp_send),
	  .ctrl.mpa_len = htons(sizeof(struct iwarp_send) - 2),
	  .ctrl.ddp_rdmap_ctrl = DDP_FLAG_LAST | cpu_to_be16(DDP_VERSION << 8) |
				 cpu_to_be16(RDMAP_VERSION << 6) |
				 cpu_to_be16(RDMAP_SEND),
	  .rx_data = siw_proc_send },
	{ /* RDMAP_SEND_INVAL */
	  .hdr_len = sizeof(struct iwarp_send_inv),
	  .ctrl.mpa_len = htons(sizeof(struct iwarp_send_inv) - 2),
	  .ctrl.ddp_rdmap_ctrl = DDP_FLAG_LAST | cpu_to_be16(DDP_VERSION << 8) |
				 cpu_to_be16(RDMAP_VERSION << 6) |
				 cpu_to_be16(RDMAP_SEND_INVAL),
	  .rx_data = siw_proc_send },
	{ /* RDMAP_SEND_SE */
	  .hdr_len = sizeof(struct iwarp_send),
	  .ctrl.mpa_len = htons(sizeof(struct iwarp_send) - 2),
	  .ctrl.ddp_rdmap_ctrl = DDP_FLAG_LAST | cpu_to_be16(DDP_VERSION << 8) |
				 cpu_to_be16(RDMAP_VERSION << 6) |
				 cpu_to_be16(RDMAP_SEND_SE),
	  .rx_data = siw_proc_send },
	{ /* RDMAP_SEND_SE_INVAL */
	  .hdr_len = sizeof(struct iwarp_send_inv),
	  .ctrl.mpa_len = htons(sizeof(struct iwarp_send_inv) - 2),
	  .ctrl.ddp_rdmap_ctrl = DDP_FLAG_LAST | cpu_to_be16(DDP_VERSION << 8) |
				 cpu_to_be16(RDMAP_VERSION << 6) |
				 cpu_to_be16(RDMAP_SEND_SE_INVAL),
	  .rx_data = siw_proc_send },
	{ /* RDMAP_TERMINATE */
	  .hdr_len = sizeof(struct iwarp_terminate),
	  .ctrl.mpa_len = htons(sizeof(struct iwarp_terminate) - 2),
	  .ctrl.ddp_rdmap_ctrl = DDP_FLAG_LAST | cpu_to_be16(DDP_VERSION << 8) |
				 cpu_to_be16(RDMAP_VERSION << 6) |
				 cpu_to_be16(RDMAP_TERMINATE),
	  .rx_data = siw_proc_terminate }
};

void siw_qp_llp_data_ready(struct sock *sk)
{
	struct siw_qp *qp;

	read_lock(&sk->sk_callback_lock);

	if (unlikely(!sk->sk_user_data || !sk_to_qp(sk)))
		goto done;

	qp = sk_to_qp(sk);

	if (likely(!qp->rx_stream.rx_suspend &&
		   down_read_trylock(&qp->state_lock))) {
		read_descriptor_t rd_desc = { .arg.data = qp, .count = 1 };

		if (likely(qp->attrs.state == SIW_QP_STATE_RTS))
			/*
			 * Implements data receive operation during
			 * socket callback. TCP gracefully catches
			 * the case where there is nothing to receive
			 * (not calling siw_tcp_rx_data() then).
			 */
			tcp_read_sock(sk, &rd_desc, siw_tcp_rx_data);

		up_read(&qp->state_lock);
	} else {
		siw_dbg_qp(qp, "unable to process RX, suspend: %d\n",
			   qp->rx_stream.rx_suspend);
	}
done:
	read_unlock(&sk->sk_callback_lock);
}

void siw_qp_llp_close(struct siw_qp *qp)
{
	siw_dbg_qp(qp, "enter llp close, state = %s\n",
		   siw_qp_state_to_string[qp->attrs.state]);

	down_write(&qp->state_lock);

	qp->rx_stream.rx_suspend = 1;
	qp->tx_ctx.tx_suspend = 1;
	qp->attrs.sk = NULL;

	switch (qp->attrs.state) {
	case SIW_QP_STATE_RTS:
	case SIW_QP_STATE_RTR:
	case SIW_QP_STATE_IDLE:
	case SIW_QP_STATE_TERMINATE:
		qp->attrs.state = SIW_QP_STATE_ERROR;
		break;
	/*
	 * SIW_QP_STATE_CLOSING:
	 *
	 * This is a forced close. shall the QP be moved to
	 * ERROR or IDLE ?
	 */
	case SIW_QP_STATE_CLOSING:
		if (tx_wqe(qp)->wr_status == SIW_WR_IDLE)
			qp->attrs.state = SIW_QP_STATE_ERROR;
		else
			qp->attrs.state = SIW_QP_STATE_IDLE;
		break;

	default:
		siw_dbg_qp(qp, "llp close: no state transition needed: %s\n",
			   siw_qp_state_to_string[qp->attrs.state]);
		break;
	}
	siw_sq_flush(qp);
	siw_rq_flush(qp);

	/*
	 * Dereference closing CEP
	 */
	if (qp->cep) {
		siw_cep_put(qp->cep);
		qp->cep = NULL;
	}

	up_write(&qp->state_lock);

	siw_dbg_qp(qp, "llp close exit: state %s\n",
		   siw_qp_state_to_string[qp->attrs.state]);
}

/*
 * socket callback routine informing about newly available send space.
 * Function schedules SQ work for processing SQ items.
 */
void siw_qp_llp_write_space(struct sock *sk)
{
	struct siw_cep *cep;

	read_lock(&sk->sk_callback_lock);

	cep  = sk_to_cep(sk);
	if (cep) {
		cep->sk_write_space(sk);

		if (!test_bit(SOCK_NOSPACE, &sk->sk_socket->flags))
			(void)siw_sq_start(cep->qp);
	}

	read_unlock(&sk->sk_callback_lock);
}

static int siw_qp_readq_init(struct siw_qp *qp, int irq_size, int orq_size)
{
	if (irq_size) {
		irq_size = roundup_pow_of_two(irq_size);
		qp->irq = vzalloc(irq_size * sizeof(struct siw_sqe));
		if (!qp->irq) {
			qp->attrs.irq_size = 0;
			return -ENOMEM;
		}
	}
	if (orq_size) {
		orq_size = roundup_pow_of_two(orq_size);
		qp->orq = vzalloc(orq_size * sizeof(struct siw_sqe));
		if (!qp->orq) {
			qp->attrs.orq_size = 0;
			qp->attrs.irq_size = 0;
			vfree(qp->irq);
			return -ENOMEM;
		}
	}
	qp->attrs.irq_size = irq_size;
	qp->attrs.orq_size = orq_size;
	siw_dbg_qp(qp, "ORD %d, IRD %d\n", orq_size, irq_size);
	return 0;
}

static int siw_qp_enable_crc(struct siw_qp *qp)
{
	struct siw_rx_stream *c_rx = &qp->rx_stream;
	struct siw_iwarp_tx *c_tx = &qp->tx_ctx;
	int size;

	if (siw_crypto_shash == NULL)
		return -ENOENT;

	size = crypto_shash_descsize(siw_crypto_shash) +
		sizeof(struct shash_desc);

	c_tx->mpa_crc_hd = kzalloc(size, GFP_KERNEL);
	c_rx->mpa_crc_hd = kzalloc(size, GFP_KERNEL);
	if (!c_tx->mpa_crc_hd || !c_rx->mpa_crc_hd) {
		kfree(c_tx->mpa_crc_hd);
		kfree(c_rx->mpa_crc_hd);
		c_tx->mpa_crc_hd = NULL;
		c_rx->mpa_crc_hd = NULL;
		return -ENOMEM;
	}
	c_tx->mpa_crc_hd->tfm = siw_crypto_shash;
	c_rx->mpa_crc_hd->tfm = siw_crypto_shash;

	return 0;
}

/*
 * Send a non signalled READ or WRITE to peer side as negotiated
 * with MPAv2 P2P setup protocol. The work request is only created
 * as a current active WR and does not consume Send Queue space.
 *
 * Caller must hold QP state lock.
 */
int siw_qp_mpa_rts(struct siw_qp *qp, enum mpa_v2_ctrl ctrl)
{
	struct siw_wqe *wqe = tx_wqe(qp);
	unsigned long flags;
	int rv = 0;

	spin_lock_irqsave(&qp->sq_lock, flags);

	if (unlikely(wqe->wr_status != SIW_WR_IDLE)) {
		spin_unlock_irqrestore(&qp->sq_lock, flags);
		return -EIO;
	}
	memset(wqe->mem, 0, sizeof(*wqe->mem) * SIW_MAX_SGE);

	wqe->wr_status = SIW_WR_QUEUED;
	wqe->sqe.flags = 0;
	wqe->sqe.num_sge = 1;
	wqe->sqe.sge[0].length = 0;
	wqe->sqe.sge[0].laddr = 0;
	wqe->sqe.sge[0].lkey = 0;
	/*
	 * While it must not be checked for inbound zero length
	 * READ/WRITE, some HW may treat STag 0 special.
	 */
	wqe->sqe.rkey = 1;
	wqe->sqe.raddr = 0;
	wqe->processed = 0;

	if (ctrl & MPA_V2_RDMA_WRITE_RTR)
		wqe->sqe.opcode = SIW_OP_WRITE;
	else if (ctrl & MPA_V2_RDMA_READ_RTR) {
		struct siw_sqe *rreq = NULL;

		wqe->sqe.opcode = SIW_OP_READ;

		spin_lock(&qp->orq_lock);

		if (qp->attrs.orq_size)
			rreq = orq_get_free(qp);
		if (rreq) {
			siw_read_to_orq(rreq, &wqe->sqe);
			qp->orq_put++;
		} else
			rv = -EIO;

		spin_unlock(&qp->orq_lock);
	} else
		rv = -EINVAL;

	if (rv)
		wqe->wr_status = SIW_WR_IDLE;

	spin_unlock_irqrestore(&qp->sq_lock, flags);

	if (!rv)
		rv = siw_sq_start(qp);

	return rv;
}

/*
 * Map memory access error to DDP tagged error
 */
enum ddp_ecode siw_tagged_error(enum siw_access_state state)
{
	switch (state) {
	case E_STAG_INVALID:
		return DDP_ECODE_T_INVALID_STAG;
	case E_BASE_BOUNDS:
		return DDP_ECODE_T_BASE_BOUNDS;
	case E_PD_MISMATCH:
		return DDP_ECODE_T_STAG_NOT_ASSOC;
	case E_ACCESS_PERM:
		/*
		 * RFC 5041 (DDP) lacks an ecode for insufficient access
		 * permissions. 'Invalid STag' seem to be the closest
		 * match though.
		 */
		return DDP_ECODE_T_INVALID_STAG;
	default:
		WARN_ON(1);
		return DDP_ECODE_T_INVALID_STAG;
	}
}

/*
 * Map memory access error to RDMAP protection error
 */
enum rdmap_ecode siw_rdmap_error(enum siw_access_state state)
{
	switch (state) {
	case E_STAG_INVALID:
		return RDMAP_ECODE_INVALID_STAG;
	case E_BASE_BOUNDS:
		return RDMAP_ECODE_BASE_BOUNDS;
	case E_PD_MISMATCH:
		return RDMAP_ECODE_STAG_NOT_ASSOC;
	case E_ACCESS_PERM:
		return RDMAP_ECODE_ACCESS_RIGHTS;
	default:
		return RDMAP_ECODE_UNSPECIFIED;
	}
}

void siw_init_terminate(struct siw_qp *qp, enum term_elayer layer, u8 etype,
			u8 ecode, int in_tx)
{
	if (!qp->term_info.valid) {
		memset(&qp->term_info, 0, sizeof(qp->term_info));
		qp->term_info.layer = layer;
		qp->term_info.etype = etype;
		qp->term_info.ecode = ecode;
		qp->term_info.in_tx = in_tx;
		qp->term_info.valid = 1;
	}
	siw_dbg_qp(qp, "init TERM: layer %d, type %d, code %d, in tx %s\n",
		   layer, etype, ecode, in_tx ? "yes" : "no");
}

/*
 * Send a TERMINATE message, as defined in RFC's 5040/5041/5044/6581.
 * Sending TERMINATE messages is best effort - such messages
 * can only be send if the QP is still connected and it does
 * not have another outbound message in-progress, i.e. the
 * TERMINATE message must not interfer with an incomplete current
 * transmit operation.
 */
void siw_send_terminate(struct siw_qp *qp)
{
	struct kvec iov[3];
	struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_EOR };
	struct iwarp_terminate *term = NULL;
	union iwarp_hdr *err_hdr = NULL;
	struct socket *s = qp->attrs.sk;
	struct siw_rx_stream *srx = &qp->rx_stream;
	union iwarp_hdr *rx_hdr = &srx->hdr;
	u32 crc = 0;
	int num_frags, len_terminate, rv;

	if (!qp->term_info.valid)
		return;

	qp->term_info.valid = 0;

	if (tx_wqe(qp)->wr_status == SIW_WR_INPROGRESS) {
		siw_dbg_qp(qp, "cannot send TERMINATE: op %d in progress\n",
			   tx_type(tx_wqe(qp)));
		return;
	}
	if (!s && qp->cep)
		/* QP not yet in RTS. Take socket from connection end point */
		s = qp->cep->sock;

	if (!s) {
		siw_dbg_qp(qp, "cannot send TERMINATE: not connected\n");
		return;
	}

	term = kzalloc(sizeof(*term), GFP_KERNEL);
	if (!term)
		return;

	term->ddp_qn = cpu_to_be32(RDMAP_UNTAGGED_QN_TERMINATE);
	term->ddp_mo = 0;
	term->ddp_msn = cpu_to_be32(1);

	iov[0].iov_base = term;
	iov[0].iov_len = sizeof(*term);

	if ((qp->term_info.layer == TERM_ERROR_LAYER_DDP) ||
	    ((qp->term_info.layer == TERM_ERROR_LAYER_RDMAP) &&
	     (qp->term_info.etype != RDMAP_ETYPE_CATASTROPHIC))) {
		err_hdr = kzalloc(sizeof(*err_hdr), GFP_KERNEL);
		if (!err_hdr) {
			kfree(term);
			return;
		}
	}
	memcpy(&term->ctrl, &iwarp_pktinfo[RDMAP_TERMINATE].ctrl,
	       sizeof(struct iwarp_ctrl));

	__rdmap_term_set_layer(term, qp->term_info.layer);
	__rdmap_term_set_etype(term, qp->term_info.etype);
	__rdmap_term_set_ecode(term, qp->term_info.ecode);

	switch (qp->term_info.layer) {
	case TERM_ERROR_LAYER_RDMAP:
		if (qp->term_info.etype == RDMAP_ETYPE_CATASTROPHIC)
			/* No additional DDP/RDMAP header to be included */
			break;

		if (qp->term_info.etype == RDMAP_ETYPE_REMOTE_PROTECTION) {
			/*
			 * Complete RDMAP frame will get attached, and
			 * DDP segment length is valid
			 */
			term->flag_m = 1;
			term->flag_d = 1;
			term->flag_r = 1;

			if (qp->term_info.in_tx) {
				struct iwarp_rdma_rreq *rreq;
				struct siw_wqe *wqe = tx_wqe(qp);

				/* Inbound RREQ error, detected during
				 * RRESP creation. Take state from
				 * current TX work queue element to
				 * reconstruct peers RREQ.
				 */
				rreq = (struct iwarp_rdma_rreq *)err_hdr;

				memcpy(&rreq->ctrl,
				       &iwarp_pktinfo[RDMAP_RDMA_READ_REQ].ctrl,
				       sizeof(struct iwarp_ctrl));

				rreq->rsvd = 0;
				rreq->ddp_qn =
					htonl(RDMAP_UNTAGGED_QN_RDMA_READ);

				/* Provide RREQ's MSN as kept aside */
				rreq->ddp_msn = htonl(wqe->sqe.sge[0].length);

				rreq->ddp_mo = htonl(wqe->processed);
				rreq->sink_stag = htonl(wqe->sqe.rkey);
				rreq->sink_to = cpu_to_be64(wqe->sqe.raddr);
				rreq->read_size = htonl(wqe->sqe.sge[0].length);
				rreq->source_stag = htonl(wqe->sqe.sge[0].lkey);
				rreq->source_to =
					cpu_to_be64(wqe->sqe.sge[0].laddr);

				iov[1].iov_base = rreq;
				iov[1].iov_len = sizeof(*rreq);

				rx_hdr = (union iwarp_hdr *)rreq;
			} else {
				/* Take RDMAP/DDP information from
				 * current (failed) inbound frame.
				 */
				iov[1].iov_base = rx_hdr;

				if (__rdmap_get_opcode(&rx_hdr->ctrl) ==
				    RDMAP_RDMA_READ_REQ)
					iov[1].iov_len =
						sizeof(struct iwarp_rdma_rreq);
				else /* SEND type */
					iov[1].iov_len =
						sizeof(struct iwarp_send);
			}
		} else {
			/* Do not report DDP hdr information if packet
			 * layout is unknown
			 */
			if ((qp->term_info.ecode == RDMAP_ECODE_VERSION) ||
			    (qp->term_info.ecode == RDMAP_ECODE_OPCODE))
				break;

			iov[1].iov_base = rx_hdr;

			/* Only DDP frame will get attached */
			if (rx_hdr->ctrl.ddp_rdmap_ctrl & DDP_FLAG_TAGGED)
				iov[1].iov_len =
					sizeof(struct iwarp_rdma_write);
			else
				iov[1].iov_len = sizeof(struct iwarp_send);

			term->flag_m = 1;
			term->flag_d = 1;
		}
		term->ctrl.mpa_len = cpu_to_be16(iov[1].iov_len);
		break;

	case TERM_ERROR_LAYER_DDP:
		/* Report error encountered while DDP processing.
		 * This can only happen as a result of inbound
		 * DDP processing
		 */

		/* Do not report DDP hdr information if packet
		 * layout is unknown
		 */
		if (((qp->term_info.etype == DDP_ETYPE_TAGGED_BUF) &&
		     (qp->term_info.ecode == DDP_ECODE_T_VERSION)) ||
		    ((qp->term_info.etype == DDP_ETYPE_UNTAGGED_BUF) &&
		     (qp->term_info.ecode == DDP_ECODE_UT_VERSION)))
			break;

		iov[1].iov_base = rx_hdr;

		if (rx_hdr->ctrl.ddp_rdmap_ctrl & DDP_FLAG_TAGGED)
			iov[1].iov_len = sizeof(struct iwarp_ctrl_tagged);
		else
			iov[1].iov_len = sizeof(struct iwarp_ctrl_untagged);

		term->flag_m = 1;
		term->flag_d = 1;
		break;

	default:
		break;
	}
	if (term->flag_m || term->flag_d || term->flag_r) {
		iov[2].iov_base = &crc;
		iov[2].iov_len = sizeof(crc);
		len_terminate = sizeof(*term) + iov[1].iov_len + MPA_CRC_SIZE;
		num_frags = 3;
	} else {
		iov[1].iov_base = &crc;
		iov[1].iov_len = sizeof(crc);
		len_terminate = sizeof(*term) + MPA_CRC_SIZE;
		num_frags = 2;
	}

	/* Adjust DDP Segment Length parameter, if valid */
	if (term->flag_m) {
		u32 real_ddp_len = be16_to_cpu(rx_hdr->ctrl.mpa_len);
		enum rdma_opcode op = __rdmap_get_opcode(&rx_hdr->ctrl);

		real_ddp_len -= iwarp_pktinfo[op].hdr_len - MPA_HDR_SIZE;
		rx_hdr->ctrl.mpa_len = cpu_to_be16(real_ddp_len);
	}

	term->ctrl.mpa_len =
		cpu_to_be16(len_terminate - (MPA_HDR_SIZE + MPA_CRC_SIZE));
	if (qp->tx_ctx.mpa_crc_hd) {
		crypto_shash_init(qp->tx_ctx.mpa_crc_hd);
		if (crypto_shash_update(qp->tx_ctx.mpa_crc_hd,
					(u8 *)iov[0].iov_base,
					iov[0].iov_len))
			goto out;

		if (num_frags == 3) {
			if (crypto_shash_update(qp->tx_ctx.mpa_crc_hd,
						(u8 *)iov[1].iov_base,
						iov[1].iov_len))
				goto out;
		}
		crypto_shash_final(qp->tx_ctx.mpa_crc_hd, (u8 *)&crc);
	}

	rv = kernel_sendmsg(s, &msg, iov, num_frags, len_terminate);
	siw_dbg_qp(qp, "sent TERM: %s, layer %d, type %d, code %d (%d bytes)\n",
		   rv == len_terminate ? "success" : "failure",
		   __rdmap_term_layer(term), __rdmap_term_etype(term),
		   __rdmap_term_ecode(term), rv);
out:
	kfree(term);
	kfree(err_hdr);
}

/*
 * Handle all attrs other than state
 */
static void siw_qp_modify_nonstate(struct siw_qp *qp,
				   struct siw_qp_attrs *attrs,
				   enum siw_qp_attr_mask mask)
{
	if (mask & SIW_QP_ATTR_ACCESS_FLAGS) {
		if (attrs->flags & SIW_RDMA_BIND_ENABLED)
			qp->attrs.flags |= SIW_RDMA_BIND_ENABLED;
		else
			qp->attrs.flags &= ~SIW_RDMA_BIND_ENABLED;

		if (attrs->flags & SIW_RDMA_WRITE_ENABLED)
			qp->attrs.flags |= SIW_RDMA_WRITE_ENABLED;
		else
			qp->attrs.flags &= ~SIW_RDMA_WRITE_ENABLED;

		if (attrs->flags & SIW_RDMA_READ_ENABLED)
			qp->attrs.flags |= SIW_RDMA_READ_ENABLED;
		else
			qp->attrs.flags &= ~SIW_RDMA_READ_ENABLED;
	}
}

static int siw_qp_nextstate_from_idle(struct siw_qp *qp,
				      struct siw_qp_attrs *attrs,
				      enum siw_qp_attr_mask mask)
{
	int rv = 0;

	switch (attrs->state) {
	case SIW_QP_STATE_RTS:
		if (attrs->flags & SIW_MPA_CRC) {
			rv = siw_qp_enable_crc(qp);
			if (rv)
				break;
		}
		if (!(mask & SIW_QP_ATTR_LLP_HANDLE)) {
			siw_dbg_qp(qp, "no socket\n");
			rv = -EINVAL;
			break;
		}
		if (!(mask & SIW_QP_ATTR_MPA)) {
			siw_dbg_qp(qp, "no MPA\n");
			rv = -EINVAL;
			break;
		}
		/*
		 * Initialize iWARP TX state
		 */
		qp->tx_ctx.ddp_msn[RDMAP_UNTAGGED_QN_SEND] = 0;
		qp->tx_ctx.ddp_msn[RDMAP_UNTAGGED_QN_RDMA_READ] = 0;
		qp->tx_ctx.ddp_msn[RDMAP_UNTAGGED_QN_TERMINATE] = 0;

		/*
		 * Initialize iWARP RX state
		 */
		qp->rx_stream.ddp_msn[RDMAP_UNTAGGED_QN_SEND] = 1;
		qp->rx_stream.ddp_msn[RDMAP_UNTAGGED_QN_RDMA_READ] = 1;
		qp->rx_stream.ddp_msn[RDMAP_UNTAGGED_QN_TERMINATE] = 1;

		/*
		 * init IRD free queue, caller has already checked
		 * limits.
		 */
		rv = siw_qp_readq_init(qp, attrs->irq_size,
				       attrs->orq_size);
		if (rv)
			break;

		qp->attrs.sk = attrs->sk;
		qp->attrs.state = SIW_QP_STATE_RTS;

		siw_dbg_qp(qp, "enter RTS: crc=%s, ord=%u, ird=%u\n",
			   attrs->flags & SIW_MPA_CRC ? "y" : "n",
			   qp->attrs.orq_size, qp->attrs.irq_size);
		break;

	case SIW_QP_STATE_ERROR:
		siw_rq_flush(qp);
		qp->attrs.state = SIW_QP_STATE_ERROR;
		if (qp->cep) {
			siw_cep_put(qp->cep);
			qp->cep = NULL;
		}
		break;

	default:
		break;
	}
	return rv;
}

static int siw_qp_nextstate_from_rts(struct siw_qp *qp,
				     struct siw_qp_attrs *attrs)
{
	int drop_conn = 0;

	switch (attrs->state) {
	case SIW_QP_STATE_CLOSING:
		/*
		 * Verbs: move to IDLE if SQ and ORQ are empty.
		 * Move to ERROR otherwise. But first of all we must
		 * close the connection. So we keep CLOSING or ERROR
		 * as a transient state, schedule connection drop work
		 * and wait for the socket state change upcall to
		 * come back closed.
		 */
		if (tx_wqe(qp)->wr_status == SIW_WR_IDLE) {
			qp->attrs.state = SIW_QP_STATE_CLOSING;
		} else {
			qp->attrs.state = SIW_QP_STATE_ERROR;
			siw_sq_flush(qp);
		}
		siw_rq_flush(qp);

		drop_conn = 1;
		break;

	case SIW_QP_STATE_TERMINATE:
		qp->attrs.state = SIW_QP_STATE_TERMINATE;

		siw_init_terminate(qp, TERM_ERROR_LAYER_RDMAP,
				   RDMAP_ETYPE_CATASTROPHIC,
				   RDMAP_ECODE_UNSPECIFIED, 1);
		drop_conn = 1;
		break;

	case SIW_QP_STATE_ERROR:
		/*
		 * This is an emergency close.
		 *
		 * Any in progress transmit operation will get
		 * cancelled.
		 * This will likely result in a protocol failure,
		 * if a TX operation is in transit. The caller
		 * could unconditional wait to give the current
		 * operation a chance to complete.
		 * Esp., how to handle the non-empty IRQ case?
		 * The peer was asking for data transfer at a valid
		 * point in time.
		 */
		siw_sq_flush(qp);
		siw_rq_flush(qp);
		qp->attrs.state = SIW_QP_STATE_ERROR;
		drop_conn = 1;
		break;

	default:
		break;
	}
	return drop_conn;
}

static void siw_qp_nextstate_from_term(struct siw_qp *qp,
				       struct siw_qp_attrs *attrs)
{
	switch (attrs->state) {
	case SIW_QP_STATE_ERROR:
		siw_rq_flush(qp);
		qp->attrs.state = SIW_QP_STATE_ERROR;

		if (tx_wqe(qp)->wr_status != SIW_WR_IDLE)
			siw_sq_flush(qp);
		break;

	default:
		break;
	}
}

static int siw_qp_nextstate_from_close(struct siw_qp *qp,
				       struct siw_qp_attrs *attrs)
{
	int rv = 0;

	switch (attrs->state) {
	case SIW_QP_STATE_IDLE:
		WARN_ON(tx_wqe(qp)->wr_status != SIW_WR_IDLE);
		qp->attrs.state = SIW_QP_STATE_IDLE;
		break;

	case SIW_QP_STATE_CLOSING:
		/*
		 * The LLP may already moved the QP to closing
		 * due to graceful peer close init
		 */
		break;

	case SIW_QP_STATE_ERROR:
		/*
		 * QP was moved to CLOSING by LLP event
		 * not yet seen by user.
		 */
		qp->attrs.state = SIW_QP_STATE_ERROR;

		if (tx_wqe(qp)->wr_status != SIW_WR_IDLE)
			siw_sq_flush(qp);

		siw_rq_flush(qp);
		break;

	default:
		siw_dbg_qp(qp, "state transition undefined: %s => %s\n",
			   siw_qp_state_to_string[qp->attrs.state],
			   siw_qp_state_to_string[attrs->state]);

		rv = -ECONNABORTED;
	}
	return rv;
}

/*
 * Caller must hold qp->state_lock
 */
int siw_qp_modify(struct siw_qp *qp, struct siw_qp_attrs *attrs,
		  enum siw_qp_attr_mask mask)
{
	int drop_conn = 0, rv = 0;

	if (!mask)
		return 0;

	siw_dbg_qp(qp, "state: %s => %s\n",
		   siw_qp_state_to_string[qp->attrs.state],
		   siw_qp_state_to_string[attrs->state]);

	if (mask != SIW_QP_ATTR_STATE)
		siw_qp_modify_nonstate(qp, attrs, mask);

	if (!(mask & SIW_QP_ATTR_STATE))
		return 0;

	switch (qp->attrs.state) {
	case SIW_QP_STATE_IDLE:
	case SIW_QP_STATE_RTR:
		rv = siw_qp_nextstate_from_idle(qp, attrs, mask);
		break;

	case SIW_QP_STATE_RTS:
		drop_conn = siw_qp_nextstate_from_rts(qp, attrs);
		break;

	case SIW_QP_STATE_TERMINATE:
		siw_qp_nextstate_from_term(qp, attrs);
		break;

	case SIW_QP_STATE_CLOSING:
		siw_qp_nextstate_from_close(qp, attrs);
		break;
	default:
		break;
	}
	if (drop_conn)
		siw_qp_cm_drop(qp, 0);

	return rv;
}

void siw_read_to_orq(struct siw_sqe *rreq, struct siw_sqe *sqe)
{
	rreq->id = sqe->id;
	rreq->opcode = sqe->opcode;
	rreq->sge[0].laddr = sqe->sge[0].laddr;
	rreq->sge[0].length = sqe->sge[0].length;
	rreq->sge[0].lkey = sqe->sge[0].lkey;
	rreq->sge[1].lkey = sqe->sge[1].lkey;
	rreq->flags = sqe->flags | SIW_WQE_VALID;
	rreq->num_sge = 1;
}

static int siw_activate_tx_from_sq(struct siw_qp *qp)
{
	struct siw_sqe *sqe;
	struct siw_wqe *wqe = tx_wqe(qp);
	int rv = 1;

	sqe = sq_get_next(qp);
	if (!sqe)
		return 0;

	memset(wqe->mem, 0, sizeof(*wqe->mem) * SIW_MAX_SGE);
	wqe->wr_status = SIW_WR_QUEUED;

	/* First copy SQE to kernel private memory */
	memcpy(&wqe->sqe, sqe, sizeof(*sqe));

	if (wqe->sqe.opcode >= SIW_NUM_OPCODES) {
		rv = -EINVAL;
		goto out;
	}
	if (wqe->sqe.flags & SIW_WQE_INLINE) {
		if (wqe->sqe.opcode != SIW_OP_SEND &&
		    wqe->sqe.opcode != SIW_OP_WRITE) {
			rv = -EINVAL;
			goto out;
		}
		if (wqe->sqe.sge[0].length > SIW_MAX_INLINE) {
			rv = -EINVAL;
			goto out;
		}
		wqe->sqe.sge[0].laddr = (uintptr_t)&wqe->sqe.sge[1];
		wqe->sqe.sge[0].lkey = 0;
		wqe->sqe.num_sge = 1;
	}
	if (wqe->sqe.flags & SIW_WQE_READ_FENCE) {
		/* A READ cannot be fenced */
		if (unlikely(wqe->sqe.opcode == SIW_OP_READ ||
			     wqe->sqe.opcode ==
				     SIW_OP_READ_LOCAL_INV)) {
			siw_dbg_qp(qp, "cannot fence read\n");
			rv = -EINVAL;
			goto out;
		}
		spin_lock(&qp->orq_lock);

		if (qp->attrs.orq_size && !siw_orq_empty(qp)) {
			qp->tx_ctx.orq_fence = 1;
			rv = 0;
		}
		spin_unlock(&qp->orq_lock);

	} else if (wqe->sqe.opcode == SIW_OP_READ ||
		   wqe->sqe.opcode == SIW_OP_READ_LOCAL_INV) {
		struct siw_sqe *rreq;

		if (unlikely(!qp->attrs.orq_size)) {
			/* We negotiated not to send READ req's */
			rv = -EINVAL;
			goto out;
		}
		wqe->sqe.num_sge = 1;

		spin_lock(&qp->orq_lock);

		rreq = orq_get_free(qp);
		if (rreq) {
			/*
			 * Make an immediate copy in ORQ to be ready
			 * to process loopback READ reply
			 */
			siw_read_to_orq(rreq, &wqe->sqe);
			qp->orq_put++;
		} else {
			qp->tx_ctx.orq_fence = 1;
			rv = 0;
		}
		spin_unlock(&qp->orq_lock);
	}

	/* Clear SQE, can be re-used by application */
	smp_store_mb(sqe->flags, 0);
	qp->sq_get++;
out:
	if (unlikely(rv < 0)) {
		siw_dbg_qp(qp, "error %d\n", rv);
		wqe->wr_status = SIW_WR_IDLE;
	}
	return rv;
}

/*
 * Must be called with SQ locked.
 * To avoid complete SQ starvation by constant inbound READ requests,
 * the active IRQ will not be served after qp->irq_burst, if the
 * SQ has pending work.
 */
int siw_activate_tx(struct siw_qp *qp)
{
	struct siw_sqe *irqe;
	struct siw_wqe *wqe = tx_wqe(qp);

	if (!qp->attrs.irq_size)
		return siw_activate_tx_from_sq(qp);

	irqe = &qp->irq[qp->irq_get % qp->attrs.irq_size];

	if (!(irqe->flags & SIW_WQE_VALID))
		return siw_activate_tx_from_sq(qp);

	/*
	 * Avoid local WQE processing starvation in case
	 * of constant inbound READ request stream
	 */
	if (sq_get_next(qp) && ++qp->irq_burst >= SIW_IRQ_MAXBURST_SQ_ACTIVE) {
		qp->irq_burst = 0;
		return siw_activate_tx_from_sq(qp);
	}
	memset(wqe->mem, 0, sizeof(*wqe->mem) * SIW_MAX_SGE);
	wqe->wr_status = SIW_WR_QUEUED;

	/* start READ RESPONSE */
	wqe->sqe.opcode = SIW_OP_READ_RESPONSE;
	wqe->sqe.flags = 0;
	if (irqe->num_sge) {
		wqe->sqe.num_sge = 1;
		wqe->sqe.sge[0].length = irqe->sge[0].length;
		wqe->sqe.sge[0].laddr = irqe->sge[0].laddr;
		wqe->sqe.sge[0].lkey = irqe->sge[0].lkey;
	} else {
		wqe->sqe.num_sge = 0;
	}

	/* Retain original RREQ's message sequence number for
	 * potential error reporting cases.
	 */
	wqe->sqe.sge[1].length = irqe->sge[1].length;

	wqe->sqe.rkey = irqe->rkey;
	wqe->sqe.raddr = irqe->raddr;

	wqe->processed = 0;
	qp->irq_get++;

	/* mark current IRQ entry free */
	smp_store_mb(irqe->flags, 0);

	return 1;
}

/*
 * Check if current CQ state qualifies for calling CQ completion
 * handler. Must be called with CQ lock held.
 */
static bool siw_cq_notify_now(struct siw_cq *cq, u32 flags)
{
	u32 cq_notify;

	if (!cq->base_cq.comp_handler)
		return false;

	/* Read application shared notification state */
	cq_notify = READ_ONCE(cq->notify->flags);

	if ((cq_notify & SIW_NOTIFY_NEXT_COMPLETION) ||
	    ((cq_notify & SIW_NOTIFY_SOLICITED) &&
	     (flags & SIW_WQE_SOLICITED))) {
		/*
		 * CQ notification is one-shot: Since the
		 * current CQE causes user notification,
		 * the CQ gets dis-aremd and must be re-aremd
		 * by the user for a new notification.
		 */
		WRITE_ONCE(cq->notify->flags, SIW_NOTIFY_NOT);

		return true;
	}
	return false;
}

int siw_sqe_complete(struct siw_qp *qp, struct siw_sqe *sqe, u32 bytes,
		     enum siw_wc_status status)
{
	struct siw_cq *cq = qp->scq;
	int rv = 0;

	if (cq) {
		u32 sqe_flags = sqe->flags;
		struct siw_cqe *cqe;
		u32 idx;
		unsigned long flags;

		spin_lock_irqsave(&cq->lock, flags);

		idx = cq->cq_put % cq->num_cqe;
		cqe = &cq->queue[idx];

		if (!READ_ONCE(cqe->flags)) {
			bool notify;

			cqe->id = sqe->id;
			cqe->opcode = sqe->opcode;
			cqe->status = status;
			cqe->imm_data = 0;
			cqe->bytes = bytes;

			if (rdma_is_kernel_res(&cq->base_cq.res))
				cqe->base_qp = &qp->base_qp;
			else
				cqe->qp_id = qp_id(qp);

			/* mark CQE valid for application */
			WRITE_ONCE(cqe->flags, SIW_WQE_VALID);
			/* recycle SQE */
			smp_store_mb(sqe->flags, 0);

			cq->cq_put++;
			notify = siw_cq_notify_now(cq, sqe_flags);

			spin_unlock_irqrestore(&cq->lock, flags);

			if (notify) {
				siw_dbg_cq(cq, "Call completion handler\n");
				cq->base_cq.comp_handler(&cq->base_cq,
						cq->base_cq.cq_context);
			}
		} else {
			spin_unlock_irqrestore(&cq->lock, flags);
			rv = -ENOMEM;
			siw_cq_event(cq, IB_EVENT_CQ_ERR);
		}
	} else {
		/* recycle SQE */
		smp_store_mb(sqe->flags, 0);
	}
	return rv;
}

int siw_rqe_complete(struct siw_qp *qp, struct siw_rqe *rqe, u32 bytes,
		     u32 inval_stag, enum siw_wc_status status)
{
	struct siw_cq *cq = qp->rcq;
	int rv = 0;

	if (cq) {
		struct siw_cqe *cqe;
		u32 idx;
		unsigned long flags;

		spin_lock_irqsave(&cq->lock, flags);

		idx = cq->cq_put % cq->num_cqe;
		cqe = &cq->queue[idx];

		if (!READ_ONCE(cqe->flags)) {
			bool notify;
			u8 cqe_flags = SIW_WQE_VALID;

			cqe->id = rqe->id;
			cqe->opcode = SIW_OP_RECEIVE;
			cqe->status = status;
			cqe->imm_data = 0;
			cqe->bytes = bytes;

			if (rdma_is_kernel_res(&cq->base_cq.res)) {
				cqe->base_qp = &qp->base_qp;
				if (inval_stag) {
					cqe_flags |= SIW_WQE_REM_INVAL;
					cqe->inval_stag = inval_stag;
				}
			} else {
				cqe->qp_id = qp_id(qp);
			}
			/* mark CQE valid for application */
			WRITE_ONCE(cqe->flags, cqe_flags);
			/* recycle RQE */
			smp_store_mb(rqe->flags, 0);

			cq->cq_put++;
			notify = siw_cq_notify_now(cq, SIW_WQE_SIGNALLED);

			spin_unlock_irqrestore(&cq->lock, flags);

			if (notify) {
				siw_dbg_cq(cq, "Call completion handler\n");
				cq->base_cq.comp_handler(&cq->base_cq,
						cq->base_cq.cq_context);
			}
		} else {
			spin_unlock_irqrestore(&cq->lock, flags);
			rv = -ENOMEM;
			siw_cq_event(cq, IB_EVENT_CQ_ERR);
		}
	} else {
		/* recycle RQE */
		smp_store_mb(rqe->flags, 0);
	}
	return rv;
}

/*
 * siw_sq_flush()
 *
 * Flush SQ and ORRQ entries to CQ.
 *
 * Must be called with QP state write lock held.
 * Therefore, SQ and ORQ lock must not be taken.
 */
void siw_sq_flush(struct siw_qp *qp)
{
	struct siw_sqe *sqe;
	struct siw_wqe *wqe = tx_wqe(qp);
	int async_event = 0;

	/*
	 * Start with completing any work currently on the ORQ
	 */
	while (qp->attrs.orq_size) {
		sqe = &qp->orq[qp->orq_get % qp->attrs.orq_size];
		if (!READ_ONCE(sqe->flags))
			break;

		if (siw_sqe_complete(qp, sqe, 0, SIW_WC_WR_FLUSH_ERR) != 0)
			break;

		WRITE_ONCE(sqe->flags, 0);
		qp->orq_get++;
	}
	/*
	 * Flush an in-progress WQE if present
	 */
	if (wqe->wr_status != SIW_WR_IDLE) {
		siw_dbg_qp(qp, "flush current SQE, type %d, status %d\n",
			   tx_type(wqe), wqe->wr_status);

		siw_wqe_put_mem(wqe, tx_type(wqe));

		if (tx_type(wqe) != SIW_OP_READ_RESPONSE &&
		    ((tx_type(wqe) != SIW_OP_READ &&
		      tx_type(wqe) != SIW_OP_READ_LOCAL_INV) ||
		     wqe->wr_status == SIW_WR_QUEUED))
			/*
			 * An in-progress Read Request is already in
			 * the ORQ
			 */
			siw_sqe_complete(qp, &wqe->sqe, wqe->bytes,
					 SIW_WC_WR_FLUSH_ERR);

		wqe->wr_status = SIW_WR_IDLE;
	}
	/*
	 * Flush the Send Queue
	 */
	while (qp->attrs.sq_size) {
		sqe = &qp->sendq[qp->sq_get % qp->attrs.sq_size];
		if (!READ_ONCE(sqe->flags))
			break;

		async_event = 1;
		if (siw_sqe_complete(qp, sqe, 0, SIW_WC_WR_FLUSH_ERR) != 0)
			/*
			 * Shall IB_EVENT_SQ_DRAINED be supressed if work
			 * completion fails?
			 */
			break;

		WRITE_ONCE(sqe->flags, 0);
		qp->sq_get++;
	}
	if (async_event)
		siw_qp_event(qp, IB_EVENT_SQ_DRAINED);
}

/*
 * siw_rq_flush()
 *
 * Flush recv queue entries to CQ. Also
 * takes care of pending active tagged and untagged
 * inbound transfers, which have target memory
 * referenced.
 *
 * Must be called with QP state write lock held.
 * Therefore, RQ lock must not be taken.
 */
void siw_rq_flush(struct siw_qp *qp)
{
	struct siw_wqe *wqe = &qp->rx_untagged.wqe_active;

	/*
	 * Flush an in-progress untagged operation if present
	 */
	if (wqe->wr_status != SIW_WR_IDLE) {
		siw_dbg_qp(qp, "flush current rqe, type %d, status %d\n",
			   rx_type(wqe), wqe->wr_status);

		siw_wqe_put_mem(wqe, rx_type(wqe));

		if (rx_type(wqe) == SIW_OP_RECEIVE) {
			siw_rqe_complete(qp, &wqe->rqe, wqe->bytes,
					 0, SIW_WC_WR_FLUSH_ERR);
		} else if (rx_type(wqe) != SIW_OP_READ &&
			   rx_type(wqe) != SIW_OP_READ_RESPONSE &&
			   rx_type(wqe) != SIW_OP_WRITE) {
			siw_sqe_complete(qp, &wqe->sqe, 0, SIW_WC_WR_FLUSH_ERR);
		}
		wqe->wr_status = SIW_WR_IDLE;
	}
	wqe = &qp->rx_tagged.wqe_active;

	if (wqe->wr_status != SIW_WR_IDLE) {
		siw_wqe_put_mem(wqe, rx_type(wqe));
		wqe->wr_status = SIW_WR_IDLE;
	}
	/*
	 * Flush the Receive Queue
	 */
	while (qp->attrs.rq_size) {
		struct siw_rqe *rqe =
			&qp->recvq[qp->rq_get % qp->attrs.rq_size];

		if (!READ_ONCE(rqe->flags))
			break;

		if (siw_rqe_complete(qp, rqe, 0, 0, SIW_WC_WR_FLUSH_ERR) != 0)
			break;

		WRITE_ONCE(rqe->flags, 0);
		qp->rq_get++;
	}
}

int siw_qp_add(struct siw_device *sdev, struct siw_qp *qp)
{
	int rv = xa_alloc(&sdev->qp_xa, &qp->base_qp.qp_num, qp, xa_limit_32b,
			  GFP_KERNEL);

	if (!rv) {
		kref_init(&qp->ref);
		qp->sdev = sdev;
		siw_dbg_qp(qp, "new QP\n");
	}
	return rv;
}

void siw_free_qp(struct kref *ref)
{
	struct siw_qp *found, *qp = container_of(ref, struct siw_qp, ref);
	struct siw_device *sdev = qp->sdev;
	unsigned long flags;

	if (qp->cep)
		siw_cep_put(qp->cep);

	found = xa_erase(&sdev->qp_xa, qp_id(qp));
	WARN_ON(found != qp);
	spin_lock_irqsave(&sdev->lock, flags);
	list_del(&qp->devq);
	spin_unlock_irqrestore(&sdev->lock, flags);

	vfree(qp->sendq);
	vfree(qp->recvq);
	vfree(qp->irq);
	vfree(qp->orq);

	siw_put_tx_cpu(qp->tx_cpu);

	atomic_dec(&sdev->num_qp);
	siw_dbg_qp(qp, "free QP\n");
	kfree_rcu(qp, rcu);
}