summaryrefslogtreecommitdiff
path: root/drivers/net/mlx5/mlx5_rxq.c
blob: 94fdd32bdfe14b141e512aeda3ab63a512432cda (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
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
/*-
 *   BSD LICENSE
 *
 *   Copyright 2015 6WIND S.A.
 *   Copyright 2015 Mellanox.
 *
 *   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.
 *     * Neither the name of 6WIND S.A. nor the names of its
 *       contributors may be used to endorse or promote products derived
 *       from this software without specific prior written permission.
 *
 *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
 *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
 *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
 *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
 *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
 *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
 *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
 *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */

#include <stddef.h>
#include <assert.h>
#include <errno.h>
#include <string.h>
#include <stdint.h>
#include <fcntl.h>

/* Verbs header. */
/* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
#ifdef PEDANTIC
#pragma GCC diagnostic ignored "-Wpedantic"
#endif
#include <infiniband/verbs.h>
#include <infiniband/arch.h>
#include <infiniband/mlx5_hw.h>
#ifdef PEDANTIC
#pragma GCC diagnostic error "-Wpedantic"
#endif

#include <rte_mbuf.h>
#include <rte_malloc.h>
#include <rte_ethdev.h>
#include <rte_common.h>
#include <rte_interrupts.h>
#include <rte_debug.h>

#include "mlx5.h"
#include "mlx5_rxtx.h"
#include "mlx5_utils.h"
#include "mlx5_autoconf.h"
#include "mlx5_defs.h"

/* Initialization data for hash RX queues. */
const struct hash_rxq_init hash_rxq_init[] = {
	[HASH_RXQ_TCPV4] = {
		.hash_fields = (IBV_EXP_RX_HASH_SRC_IPV4 |
				IBV_EXP_RX_HASH_DST_IPV4 |
				IBV_EXP_RX_HASH_SRC_PORT_TCP |
				IBV_EXP_RX_HASH_DST_PORT_TCP),
		.dpdk_rss_hf = ETH_RSS_NONFRAG_IPV4_TCP,
		.flow_priority = 0,
		.flow_spec.tcp_udp = {
			.type = IBV_EXP_FLOW_SPEC_TCP,
			.size = sizeof(hash_rxq_init[0].flow_spec.tcp_udp),
		},
		.underlayer = &hash_rxq_init[HASH_RXQ_IPV4],
	},
	[HASH_RXQ_UDPV4] = {
		.hash_fields = (IBV_EXP_RX_HASH_SRC_IPV4 |
				IBV_EXP_RX_HASH_DST_IPV4 |
				IBV_EXP_RX_HASH_SRC_PORT_UDP |
				IBV_EXP_RX_HASH_DST_PORT_UDP),
		.dpdk_rss_hf = ETH_RSS_NONFRAG_IPV4_UDP,
		.flow_priority = 0,
		.flow_spec.tcp_udp = {
			.type = IBV_EXP_FLOW_SPEC_UDP,
			.size = sizeof(hash_rxq_init[0].flow_spec.tcp_udp),
		},
		.underlayer = &hash_rxq_init[HASH_RXQ_IPV4],
	},
	[HASH_RXQ_IPV4] = {
		.hash_fields = (IBV_EXP_RX_HASH_SRC_IPV4 |
				IBV_EXP_RX_HASH_DST_IPV4),
		.dpdk_rss_hf = (ETH_RSS_IPV4 |
				ETH_RSS_FRAG_IPV4),
		.flow_priority = 1,
		.flow_spec.ipv4 = {
			.type = IBV_EXP_FLOW_SPEC_IPV4,
			.size = sizeof(hash_rxq_init[0].flow_spec.ipv4),
		},
		.underlayer = &hash_rxq_init[HASH_RXQ_ETH],
	},
	[HASH_RXQ_TCPV6] = {
		.hash_fields = (IBV_EXP_RX_HASH_SRC_IPV6 |
				IBV_EXP_RX_HASH_DST_IPV6 |
				IBV_EXP_RX_HASH_SRC_PORT_TCP |
				IBV_EXP_RX_HASH_DST_PORT_TCP),
		.dpdk_rss_hf = ETH_RSS_NONFRAG_IPV6_TCP,
		.flow_priority = 0,
		.flow_spec.tcp_udp = {
			.type = IBV_EXP_FLOW_SPEC_TCP,
			.size = sizeof(hash_rxq_init[0].flow_spec.tcp_udp),
		},
		.underlayer = &hash_rxq_init[HASH_RXQ_IPV6],
	},
	[HASH_RXQ_UDPV6] = {
		.hash_fields = (IBV_EXP_RX_HASH_SRC_IPV6 |
				IBV_EXP_RX_HASH_DST_IPV6 |
				IBV_EXP_RX_HASH_SRC_PORT_UDP |
				IBV_EXP_RX_HASH_DST_PORT_UDP),
		.dpdk_rss_hf = ETH_RSS_NONFRAG_IPV6_UDP,
		.flow_priority = 0,
		.flow_spec.tcp_udp = {
			.type = IBV_EXP_FLOW_SPEC_UDP,
			.size = sizeof(hash_rxq_init[0].flow_spec.tcp_udp),
		},
		.underlayer = &hash_rxq_init[HASH_RXQ_IPV6],
	},
	[HASH_RXQ_IPV6] = {
		.hash_fields = (IBV_EXP_RX_HASH_SRC_IPV6 |
				IBV_EXP_RX_HASH_DST_IPV6),
		.dpdk_rss_hf = (ETH_RSS_IPV6 |
				ETH_RSS_FRAG_IPV6),
		.flow_priority = 1,
		.flow_spec.ipv6 = {
			.type = IBV_EXP_FLOW_SPEC_IPV6,
			.size = sizeof(hash_rxq_init[0].flow_spec.ipv6),
		},
		.underlayer = &hash_rxq_init[HASH_RXQ_ETH],
	},
	[HASH_RXQ_ETH] = {
		.hash_fields = 0,
		.dpdk_rss_hf = 0,
		.flow_priority = 2,
		.flow_spec.eth = {
			.type = IBV_EXP_FLOW_SPEC_ETH,
			.size = sizeof(hash_rxq_init[0].flow_spec.eth),
		},
		.underlayer = NULL,
	},
};

/* Number of entries in hash_rxq_init[]. */
const unsigned int hash_rxq_init_n = RTE_DIM(hash_rxq_init);

/* Initialization data for hash RX queue indirection tables. */
static const struct ind_table_init ind_table_init[] = {
	{
		.max_size = -1u, /* Superseded by HW limitations. */
		.hash_types =
			1 << HASH_RXQ_TCPV4 |
			1 << HASH_RXQ_UDPV4 |
			1 << HASH_RXQ_IPV4 |
			1 << HASH_RXQ_TCPV6 |
			1 << HASH_RXQ_UDPV6 |
			1 << HASH_RXQ_IPV6 |
			0,
		.hash_types_n = 6,
	},
	{
		.max_size = 1,
		.hash_types = 1 << HASH_RXQ_ETH,
		.hash_types_n = 1,
	},
};

#define IND_TABLE_INIT_N RTE_DIM(ind_table_init)

/* Default RSS hash key also used for ConnectX-3. */
uint8_t rss_hash_default_key[] = {
	0x2c, 0xc6, 0x81, 0xd1,
	0x5b, 0xdb, 0xf4, 0xf7,
	0xfc, 0xa2, 0x83, 0x19,
	0xdb, 0x1a, 0x3e, 0x94,
	0x6b, 0x9e, 0x38, 0xd9,
	0x2c, 0x9c, 0x03, 0xd1,
	0xad, 0x99, 0x44, 0xa7,
	0xd9, 0x56, 0x3d, 0x59,
	0x06, 0x3c, 0x25, 0xf3,
	0xfc, 0x1f, 0xdc, 0x2a,
};

/* Length of the default RSS hash key. */
const size_t rss_hash_default_key_len = sizeof(rss_hash_default_key);

/**
 * Populate flow steering rule for a given hash RX queue type using
 * information from hash_rxq_init[]. Nothing is written to flow_attr when
 * flow_attr_size is not large enough, but the required size is still returned.
 *
 * @param priv
 *   Pointer to private structure.
 * @param[out] flow_attr
 *   Pointer to flow attribute structure to fill. Note that the allocated
 *   area must be larger and large enough to hold all flow specifications.
 * @param flow_attr_size
 *   Entire size of flow_attr and trailing room for flow specifications.
 * @param type
 *   Hash RX queue type to use for flow steering rule.
 *
 * @return
 *   Total size of the flow attribute buffer. No errors are defined.
 */
size_t
priv_flow_attr(struct priv *priv, struct ibv_exp_flow_attr *flow_attr,
	       size_t flow_attr_size, enum hash_rxq_type type)
{
	size_t offset = sizeof(*flow_attr);
	const struct hash_rxq_init *init = &hash_rxq_init[type];

	assert(priv != NULL);
	assert((size_t)type < RTE_DIM(hash_rxq_init));
	do {
		offset += init->flow_spec.hdr.size;
		init = init->underlayer;
	} while (init != NULL);
	if (offset > flow_attr_size)
		return offset;
	flow_attr_size = offset;
	init = &hash_rxq_init[type];
	*flow_attr = (struct ibv_exp_flow_attr){
		.type = IBV_EXP_FLOW_ATTR_NORMAL,
		/* Priorities < 3 are reserved for flow director. */
		.priority = init->flow_priority + 3,
		.num_of_specs = 0,
		.port = priv->port,
		.flags = 0,
	};
	do {
		offset -= init->flow_spec.hdr.size;
		memcpy((void *)((uintptr_t)flow_attr + offset),
		       &init->flow_spec,
		       init->flow_spec.hdr.size);
		++flow_attr->num_of_specs;
		init = init->underlayer;
	} while (init != NULL);
	return flow_attr_size;
}

/**
 * Convert hash type position in indirection table initializer to
 * hash RX queue type.
 *
 * @param table
 *   Indirection table initializer.
 * @param pos
 *   Hash type position.
 *
 * @return
 *   Hash RX queue type.
 */
static enum hash_rxq_type
hash_rxq_type_from_pos(const struct ind_table_init *table, unsigned int pos)
{
	enum hash_rxq_type type = HASH_RXQ_TCPV4;

	assert(pos < table->hash_types_n);
	do {
		if ((table->hash_types & (1 << type)) && (pos-- == 0))
			break;
		++type;
	} while (1);
	return type;
}

/**
 * Filter out disabled hash RX queue types from ind_table_init[].
 *
 * @param priv
 *   Pointer to private structure.
 * @param[out] table
 *   Output table.
 *
 * @return
 *   Number of table entries.
 */
static unsigned int
priv_make_ind_table_init(struct priv *priv,
			 struct ind_table_init (*table)[IND_TABLE_INIT_N])
{
	uint64_t rss_hf;
	unsigned int i;
	unsigned int j;
	unsigned int table_n = 0;
	/* Mandatory to receive frames not handled by normal hash RX queues. */
	unsigned int hash_types_sup = 1 << HASH_RXQ_ETH;

	rss_hf = priv->rss_hf;
	/* Process other protocols only if more than one queue. */
	if (priv->rxqs_n > 1)
		for (i = 0; (i != hash_rxq_init_n); ++i)
			if (rss_hf & hash_rxq_init[i].dpdk_rss_hf)
				hash_types_sup |= (1 << i);

	/* Filter out entries whose protocols are not in the set. */
	for (i = 0, j = 0; (i != IND_TABLE_INIT_N); ++i) {
		unsigned int nb;
		unsigned int h;

		/* j is increased only if the table has valid protocols. */
		assert(j <= i);
		(*table)[j] = ind_table_init[i];
		(*table)[j].hash_types &= hash_types_sup;
		for (h = 0, nb = 0; (h != hash_rxq_init_n); ++h)
			if (((*table)[j].hash_types >> h) & 0x1)
				++nb;
		(*table)[i].hash_types_n = nb;
		if (nb) {
			++table_n;
			++j;
		}
	}
	return table_n;
}

/**
 * Initialize hash RX queues and indirection table.
 *
 * @param priv
 *   Pointer to private structure.
 *
 * @return
 *   0 on success, errno value on failure.
 */
int
priv_create_hash_rxqs(struct priv *priv)
{
	struct ibv_exp_wq *wqs[priv->reta_idx_n];
	struct ind_table_init ind_table_init[IND_TABLE_INIT_N];
	unsigned int ind_tables_n =
		priv_make_ind_table_init(priv, &ind_table_init);
	unsigned int hash_rxqs_n = 0;
	struct hash_rxq (*hash_rxqs)[] = NULL;
	struct ibv_exp_rwq_ind_table *(*ind_tables)[] = NULL;
	unsigned int i;
	unsigned int j;
	unsigned int k;
	int err = 0;

	assert(priv->ind_tables == NULL);
	assert(priv->ind_tables_n == 0);
	assert(priv->hash_rxqs == NULL);
	assert(priv->hash_rxqs_n == 0);
	assert(priv->pd != NULL);
	assert(priv->ctx != NULL);
	if (priv->isolated)
		return 0;
	if (priv->rxqs_n == 0)
		return EINVAL;
	assert(priv->rxqs != NULL);
	if (ind_tables_n == 0) {
		ERROR("all hash RX queue types have been filtered out,"
		      " indirection table cannot be created");
		return EINVAL;
	}
	if (priv->rxqs_n & (priv->rxqs_n - 1)) {
		INFO("%u RX queues are configured, consider rounding this"
		     " number to the next power of two for better balancing",
		     priv->rxqs_n);
		DEBUG("indirection table extended to assume %u WQs",
		      priv->reta_idx_n);
	}
	for (i = 0; (i != priv->reta_idx_n); ++i) {
		struct rxq_ctrl *rxq_ctrl;

		rxq_ctrl = container_of((*priv->rxqs)[(*priv->reta_idx)[i]],
					struct rxq_ctrl, rxq);
		wqs[i] = rxq_ctrl->wq;
	}
	/* Get number of hash RX queues to configure. */
	for (i = 0, hash_rxqs_n = 0; (i != ind_tables_n); ++i)
		hash_rxqs_n += ind_table_init[i].hash_types_n;
	DEBUG("allocating %u hash RX queues for %u WQs, %u indirection tables",
	      hash_rxqs_n, priv->rxqs_n, ind_tables_n);
	/* Create indirection tables. */
	ind_tables = rte_calloc(__func__, ind_tables_n,
				sizeof((*ind_tables)[0]), 0);
	if (ind_tables == NULL) {
		err = ENOMEM;
		ERROR("cannot allocate indirection tables container: %s",
		      strerror(err));
		goto error;
	}
	for (i = 0; (i != ind_tables_n); ++i) {
		struct ibv_exp_rwq_ind_table_init_attr ind_init_attr = {
			.pd = priv->pd,
			.log_ind_tbl_size = 0, /* Set below. */
			.ind_tbl = wqs,
			.comp_mask = 0,
		};
		unsigned int ind_tbl_size = ind_table_init[i].max_size;
		struct ibv_exp_rwq_ind_table *ind_table;

		if (priv->reta_idx_n < ind_tbl_size)
			ind_tbl_size = priv->reta_idx_n;
		ind_init_attr.log_ind_tbl_size = log2above(ind_tbl_size);
		errno = 0;
		ind_table = ibv_exp_create_rwq_ind_table(priv->ctx,
							 &ind_init_attr);
		if (ind_table != NULL) {
			(*ind_tables)[i] = ind_table;
			continue;
		}
		/* Not clear whether errno is set. */
		err = (errno ? errno : EINVAL);
		ERROR("RX indirection table creation failed with error %d: %s",
		      err, strerror(err));
		goto error;
	}
	/* Allocate array that holds hash RX queues and related data. */
	hash_rxqs = rte_calloc(__func__, hash_rxqs_n,
			       sizeof((*hash_rxqs)[0]), 0);
	if (hash_rxqs == NULL) {
		err = ENOMEM;
		ERROR("cannot allocate hash RX queues container: %s",
		      strerror(err));
		goto error;
	}
	for (i = 0, j = 0, k = 0;
	     ((i != hash_rxqs_n) && (j != ind_tables_n));
	     ++i) {
		struct hash_rxq *hash_rxq = &(*hash_rxqs)[i];
		enum hash_rxq_type type =
			hash_rxq_type_from_pos(&ind_table_init[j], k);
		struct rte_eth_rss_conf *priv_rss_conf =
			(*priv->rss_conf)[type];
		struct ibv_exp_rx_hash_conf hash_conf = {
			.rx_hash_function = IBV_EXP_RX_HASH_FUNC_TOEPLITZ,
			.rx_hash_key_len = (priv_rss_conf ?
					    priv_rss_conf->rss_key_len :
					    rss_hash_default_key_len),
			.rx_hash_key = (priv_rss_conf ?
					priv_rss_conf->rss_key :
					rss_hash_default_key),
			.rx_hash_fields_mask = hash_rxq_init[type].hash_fields,
			.rwq_ind_tbl = (*ind_tables)[j],
		};
		struct ibv_exp_qp_init_attr qp_init_attr = {
			.max_inl_recv = 0, /* Currently not supported. */
			.qp_type = IBV_QPT_RAW_PACKET,
			.comp_mask = (IBV_EXP_QP_INIT_ATTR_PD |
				      IBV_EXP_QP_INIT_ATTR_RX_HASH),
			.pd = priv->pd,
			.rx_hash_conf = &hash_conf,
			.port_num = priv->port,
		};

		DEBUG("using indirection table %u for hash RX queue %u type %d",
		      j, i, type);
		*hash_rxq = (struct hash_rxq){
			.priv = priv,
			.qp = ibv_exp_create_qp(priv->ctx, &qp_init_attr),
			.type = type,
		};
		if (hash_rxq->qp == NULL) {
			err = (errno ? errno : EINVAL);
			ERROR("Hash RX QP creation failure: %s",
			      strerror(err));
			goto error;
		}
		if (++k < ind_table_init[j].hash_types_n)
			continue;
		/* Switch to the next indirection table and reset hash RX
		 * queue type array index. */
		++j;
		k = 0;
	}
	priv->ind_tables = ind_tables;
	priv->ind_tables_n = ind_tables_n;
	priv->hash_rxqs = hash_rxqs;
	priv->hash_rxqs_n = hash_rxqs_n;
	assert(err == 0);
	return 0;
error:
	if (hash_rxqs != NULL) {
		for (i = 0; (i != hash_rxqs_n); ++i) {
			struct ibv_qp *qp = (*hash_rxqs)[i].qp;

			if (qp == NULL)
				continue;
			claim_zero(ibv_destroy_qp(qp));
		}
		rte_free(hash_rxqs);
	}
	if (ind_tables != NULL) {
		for (j = 0; (j != ind_tables_n); ++j) {
			struct ibv_exp_rwq_ind_table *ind_table =
				(*ind_tables)[j];

			if (ind_table == NULL)
				continue;
			claim_zero(ibv_exp_destroy_rwq_ind_table(ind_table));
		}
		rte_free(ind_tables);
	}
	return err;
}

/**
 * Clean up hash RX queues and indirection table.
 *
 * @param priv
 *   Pointer to private structure.
 */
void
priv_destroy_hash_rxqs(struct priv *priv)
{
	unsigned int i;

	DEBUG("destroying %u hash RX queues", priv->hash_rxqs_n);
	if (priv->hash_rxqs_n == 0) {
		assert(priv->hash_rxqs == NULL);
		assert(priv->ind_tables == NULL);
		return;
	}
	for (i = 0; (i != priv->hash_rxqs_n); ++i) {
		struct hash_rxq *hash_rxq = &(*priv->hash_rxqs)[i];
		unsigned int j, k;

		assert(hash_rxq->priv == priv);
		assert(hash_rxq->qp != NULL);
		/* Also check that there are no remaining flows. */
		for (j = 0; (j != RTE_DIM(hash_rxq->special_flow)); ++j)
			for (k = 0;
			     (k != RTE_DIM(hash_rxq->special_flow[j]));
			     ++k)
				assert(hash_rxq->special_flow[j][k] == NULL);
		for (j = 0; (j != RTE_DIM(hash_rxq->mac_flow)); ++j)
			for (k = 0; (k != RTE_DIM(hash_rxq->mac_flow[j])); ++k)
				assert(hash_rxq->mac_flow[j][k] == NULL);
		claim_zero(ibv_destroy_qp(hash_rxq->qp));
	}
	priv->hash_rxqs_n = 0;
	rte_free(priv->hash_rxqs);
	priv->hash_rxqs = NULL;
	for (i = 0; (i != priv->ind_tables_n); ++i) {
		struct ibv_exp_rwq_ind_table *ind_table =
			(*priv->ind_tables)[i];

		assert(ind_table != NULL);
		claim_zero(ibv_exp_destroy_rwq_ind_table(ind_table));
	}
	priv->ind_tables_n = 0;
	rte_free(priv->ind_tables);
	priv->ind_tables = NULL;
}

/**
 * Check whether a given flow type is allowed.
 *
 * @param priv
 *   Pointer to private structure.
 * @param type
 *   Flow type to check.
 *
 * @return
 *   Nonzero if the given flow type is allowed.
 */
int
priv_allow_flow_type(struct priv *priv, enum hash_rxq_flow_type type)
{
	/* Only FLOW_TYPE_PROMISC is allowed when promiscuous mode
	 * has been requested. */
	if (priv->promisc_req)
		return type == HASH_RXQ_FLOW_TYPE_PROMISC;
	switch (type) {
	case HASH_RXQ_FLOW_TYPE_PROMISC:
		return !!priv->promisc_req;
	case HASH_RXQ_FLOW_TYPE_ALLMULTI:
		return !!priv->allmulti_req;
	case HASH_RXQ_FLOW_TYPE_BROADCAST:
	case HASH_RXQ_FLOW_TYPE_IPV6MULTI:
		/* If allmulti is enabled, broadcast and ipv6multi
		 * are unnecessary. */
		return !priv->allmulti_req;
	case HASH_RXQ_FLOW_TYPE_MAC:
		return 1;
	default:
		/* Unsupported flow type is not allowed. */
		return 0;
	}
	return 0;
}

/**
 * Automatically enable/disable flows according to configuration.
 *
 * @param priv
 *   Private structure.
 *
 * @return
 *   0 on success, errno value on failure.
 */
int
priv_rehash_flows(struct priv *priv)
{
	enum hash_rxq_flow_type i;

	for (i = HASH_RXQ_FLOW_TYPE_PROMISC;
			i != RTE_DIM((*priv->hash_rxqs)[0].special_flow);
			++i)
		if (!priv_allow_flow_type(priv, i)) {
			priv_special_flow_disable(priv, i);
		} else {
			int ret = priv_special_flow_enable(priv, i);

			if (ret)
				return ret;
		}
	if (priv_allow_flow_type(priv, HASH_RXQ_FLOW_TYPE_MAC))
		return priv_mac_addrs_enable(priv);
	priv_mac_addrs_disable(priv);
	return 0;
}

/**
 * Allocate RX queue elements.
 *
 * @param rxq_ctrl
 *   Pointer to RX queue structure.
 * @param elts_n
 *   Number of elements to allocate.
 *
 * @return
 *   0 on success, errno value on failure.
 */
static int
rxq_alloc_elts(struct rxq_ctrl *rxq_ctrl, unsigned int elts_n)
{
	const unsigned int sges_n = 1 << rxq_ctrl->rxq.sges_n;
	unsigned int i;
	int ret = 0;

	/* Iterate on segments. */
	for (i = 0; (i != elts_n); ++i) {
		struct rte_mbuf *buf;
		volatile struct mlx5_wqe_data_seg *scat =
			&(*rxq_ctrl->rxq.wqes)[i];

		buf = rte_pktmbuf_alloc(rxq_ctrl->rxq.mp);
		if (buf == NULL) {
			ERROR("%p: empty mbuf pool", (void *)rxq_ctrl);
			ret = ENOMEM;
			goto error;
		}
		/* Headroom is reserved by rte_pktmbuf_alloc(). */
		assert(DATA_OFF(buf) == RTE_PKTMBUF_HEADROOM);
		/* Buffer is supposed to be empty. */
		assert(rte_pktmbuf_data_len(buf) == 0);
		assert(rte_pktmbuf_pkt_len(buf) == 0);
		assert(!buf->next);
		/* Only the first segment keeps headroom. */
		if (i % sges_n)
			SET_DATA_OFF(buf, 0);
		PORT(buf) = rxq_ctrl->rxq.port_id;
		DATA_LEN(buf) = rte_pktmbuf_tailroom(buf);
		PKT_LEN(buf) = DATA_LEN(buf);
		NB_SEGS(buf) = 1;
		/* scat->addr must be able to store a pointer. */
		assert(sizeof(scat->addr) >= sizeof(uintptr_t));
		*scat = (struct mlx5_wqe_data_seg){
			.addr = htonll(rte_pktmbuf_mtod(buf, uintptr_t)),
			.byte_count = htonl(DATA_LEN(buf)),
			.lkey = htonl(rxq_ctrl->mr->lkey),
		};
		(*rxq_ctrl->rxq.elts)[i] = buf;
	}
	if (rxq_check_vec_support(&rxq_ctrl->rxq) > 0) {
		struct rxq *rxq = &rxq_ctrl->rxq;
		struct rte_mbuf *mbuf_init = &rxq->fake_mbuf;

		assert(rxq->elts_n == rxq->cqe_n);
		/* Initialize default rearm_data for vPMD. */
		mbuf_init->data_off = RTE_PKTMBUF_HEADROOM;
		rte_mbuf_refcnt_set(mbuf_init, 1);
		mbuf_init->nb_segs = 1;
		mbuf_init->port = rxq->port_id;
		/*
		 * prevent compiler reordering:
		 * rearm_data covers previous fields.
		 */
		rte_compiler_barrier();
		rxq->mbuf_initializer = *(uint64_t *)&mbuf_init->rearm_data;
		/* Padding with a fake mbuf for vectorized Rx. */
		for (i = 0; i < MLX5_VPMD_DESCS_PER_LOOP; ++i)
			(*rxq->elts)[elts_n + i] = &rxq->fake_mbuf;
	}
	DEBUG("%p: allocated and configured %u segments (max %u packets)",
	      (void *)rxq_ctrl, elts_n, elts_n / (1 << rxq_ctrl->rxq.sges_n));
	assert(ret == 0);
	return 0;
error:
	elts_n = i;
	for (i = 0; (i != elts_n); ++i) {
		if ((*rxq_ctrl->rxq.elts)[i] != NULL)
			rte_pktmbuf_free_seg((*rxq_ctrl->rxq.elts)[i]);
		(*rxq_ctrl->rxq.elts)[i] = NULL;
	}
	DEBUG("%p: failed, freed everything", (void *)rxq_ctrl);
	assert(ret > 0);
	return ret;
}

/**
 * Free RX queue elements.
 *
 * @param rxq_ctrl
 *   Pointer to RX queue structure.
 */
static void
rxq_free_elts(struct rxq_ctrl *rxq_ctrl)
{
	struct rxq *rxq = &rxq_ctrl->rxq;
	const uint16_t q_n = (1 << rxq->elts_n);
	const uint16_t q_mask = q_n - 1;
	uint16_t used = q_n - (rxq->rq_ci - rxq->rq_pi);
	uint16_t i;

	DEBUG("%p: freeing WRs", (void *)rxq_ctrl);
	if (rxq->elts == NULL)
		return;
	/**
	 * Some mbuf in the Ring belongs to the application.  They cannot be
	 * freed.
	 */
	if (rxq_check_vec_support(rxq) > 0) {
		for (i = 0; i < used; ++i)
			(*rxq->elts)[(rxq->rq_ci + i) & q_mask] = NULL;
		rxq->rq_pi = rxq->rq_ci;
	}
	for (i = 0; (i != (1u << rxq->elts_n)); ++i) {
		if ((*rxq->elts)[i] != NULL)
			rte_pktmbuf_free_seg((*rxq->elts)[i]);
		(*rxq->elts)[i] = NULL;
	}
}

/**
 * Clean up a RX queue.
 *
 * Destroy objects, free allocated memory and reset the structure for reuse.
 *
 * @param rxq_ctrl
 *   Pointer to RX queue structure.
 */
void
rxq_cleanup(struct rxq_ctrl *rxq_ctrl)
{
	DEBUG("cleaning up %p", (void *)rxq_ctrl);
	rxq_free_elts(rxq_ctrl);
	if (rxq_ctrl->fdir_queue != NULL)
		priv_fdir_queue_destroy(rxq_ctrl->priv, rxq_ctrl->fdir_queue);
	if (rxq_ctrl->wq != NULL)
		claim_zero(ibv_exp_destroy_wq(rxq_ctrl->wq));
	if (rxq_ctrl->cq != NULL)
		claim_zero(ibv_destroy_cq(rxq_ctrl->cq));
	if (rxq_ctrl->channel != NULL)
		claim_zero(ibv_destroy_comp_channel(rxq_ctrl->channel));
	if (rxq_ctrl->mr != NULL)
		claim_zero(ibv_dereg_mr(rxq_ctrl->mr));
	memset(rxq_ctrl, 0, sizeof(*rxq_ctrl));
}

/**
 * Initialize RX queue.
 *
 * @param tmpl
 *   Pointer to RX queue control template.
 *
 * @return
 *   0 on success, errno value on failure.
 */
static inline int
rxq_setup(struct rxq_ctrl *tmpl)
{
	struct ibv_cq *ibcq = tmpl->cq;
	struct ibv_mlx5_cq_info cq_info;
	struct mlx5_rwq *rwq = container_of(tmpl->wq, struct mlx5_rwq, wq);
	const uint16_t desc_n =
		(1 << tmpl->rxq.elts_n) + tmpl->priv->rx_vec_en *
		MLX5_VPMD_DESCS_PER_LOOP;
	struct rte_mbuf *(*elts)[desc_n] =
		rte_calloc_socket("RXQ", 1, sizeof(*elts), 0, tmpl->socket);
	if (ibv_mlx5_exp_get_cq_info(ibcq, &cq_info)) {
		ERROR("Unable to query CQ info. check your OFED.");
		return ENOTSUP;
	}
	if (cq_info.cqe_size != RTE_CACHE_LINE_SIZE) {
		ERROR("Wrong MLX5_CQE_SIZE environment variable value: "
		      "it should be set to %u", RTE_CACHE_LINE_SIZE);
		return EINVAL;
	}
	if (elts == NULL)
		return ENOMEM;
	tmpl->rxq.rq_db = rwq->rq.db;
	tmpl->rxq.cqe_n = log2above(cq_info.cqe_cnt);
	tmpl->rxq.cq_ci = 0;
	tmpl->rxq.rq_ci = 0;
	tmpl->rxq.rq_pi = 0;
	tmpl->rxq.cq_db = cq_info.dbrec;
	tmpl->rxq.wqes =
		(volatile struct mlx5_wqe_data_seg (*)[])
		(uintptr_t)rwq->rq.buff;
	tmpl->rxq.cqes =
		(volatile struct mlx5_cqe (*)[])
		(uintptr_t)cq_info.buf;
	tmpl->rxq.elts = elts;
	return 0;
}

/**
 * Configure a RX queue.
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 * @param rxq_ctrl
 *   Pointer to RX queue structure.
 * @param desc
 *   Number of descriptors to configure in queue.
 * @param socket
 *   NUMA socket on which memory must be allocated.
 * @param[in] conf
 *   Thresholds parameters.
 * @param mp
 *   Memory pool for buffer allocations.
 *
 * @return
 *   0 on success, errno value on failure.
 */
static int
rxq_ctrl_setup(struct rte_eth_dev *dev, struct rxq_ctrl *rxq_ctrl,
	       uint16_t desc, unsigned int socket,
	       const struct rte_eth_rxconf *conf, struct rte_mempool *mp)
{
	struct priv *priv = dev->data->dev_private;
	struct rxq_ctrl tmpl = {
		.priv = priv,
		.socket = socket,
		.rxq = {
			.elts_n = log2above(desc),
			.mp = mp,
			.rss_hash = priv->rxqs_n > 1,
		},
	};
	struct ibv_exp_wq_attr mod;
	union {
		struct ibv_exp_cq_init_attr cq;
		struct ibv_exp_wq_init_attr wq;
		struct ibv_exp_cq_attr cq_attr;
	} attr;
	unsigned int mb_len = rte_pktmbuf_data_room_size(mp);
	unsigned int cqe_n = desc - 1;
	const uint16_t desc_n =
		desc + priv->rx_vec_en * MLX5_VPMD_DESCS_PER_LOOP;
	struct rte_mbuf *(*elts)[desc_n] = NULL;
	int ret = 0;

	(void)conf; /* Thresholds configuration (ignored). */
	/* Enable scattered packets support for this queue if necessary. */
	assert(mb_len >= RTE_PKTMBUF_HEADROOM);
	if (dev->data->dev_conf.rxmode.max_rx_pkt_len <=
	    (mb_len - RTE_PKTMBUF_HEADROOM)) {
		tmpl.rxq.sges_n = 0;
	} else if (dev->data->dev_conf.rxmode.enable_scatter) {
		unsigned int size =
			RTE_PKTMBUF_HEADROOM +
			dev->data->dev_conf.rxmode.max_rx_pkt_len;
		unsigned int sges_n;

		/*
		 * Determine the number of SGEs needed for a full packet
		 * and round it to the next power of two.
		 */
		sges_n = log2above((size / mb_len) + !!(size % mb_len));
		tmpl.rxq.sges_n = sges_n;
		/* Make sure rxq.sges_n did not overflow. */
		size = mb_len * (1 << tmpl.rxq.sges_n);
		size -= RTE_PKTMBUF_HEADROOM;
		if (size < dev->data->dev_conf.rxmode.max_rx_pkt_len) {
			ERROR("%p: too many SGEs (%u) needed to handle"
			      " requested maximum packet size %u",
			      (void *)dev,
			      1 << sges_n,
			      dev->data->dev_conf.rxmode.max_rx_pkt_len);
			return EOVERFLOW;
		}
	} else {
		WARN("%p: the requested maximum Rx packet size (%u) is"
		     " larger than a single mbuf (%u) and scattered"
		     " mode has not been requested",
		     (void *)dev,
		     dev->data->dev_conf.rxmode.max_rx_pkt_len,
		     mb_len - RTE_PKTMBUF_HEADROOM);
	}
	DEBUG("%p: maximum number of segments per packet: %u",
	      (void *)dev, 1 << tmpl.rxq.sges_n);
	if (desc % (1 << tmpl.rxq.sges_n)) {
		ERROR("%p: number of RX queue descriptors (%u) is not a"
		      " multiple of SGEs per packet (%u)",
		      (void *)dev,
		      desc,
		      1 << tmpl.rxq.sges_n);
		return EINVAL;
	}
	/* Toggle RX checksum offload if hardware supports it. */
	if (priv->hw_csum)
		tmpl.rxq.csum = !!dev->data->dev_conf.rxmode.hw_ip_checksum;
	if (priv->hw_csum_l2tun)
		tmpl.rxq.csum_l2tun =
			!!dev->data->dev_conf.rxmode.hw_ip_checksum;
	/* Use the entire RX mempool as the memory region. */
	tmpl.mr = mlx5_mp2mr(priv->pd, mp);
	if (tmpl.mr == NULL) {
		ret = EINVAL;
		ERROR("%p: MR creation failure: %s",
		      (void *)dev, strerror(ret));
		goto error;
	}
	if (dev->data->dev_conf.intr_conf.rxq) {
		tmpl.channel = ibv_create_comp_channel(priv->ctx);
		if (tmpl.channel == NULL) {
			ret = ENOMEM;
			ERROR("%p: Rx interrupt completion channel creation"
			      " failure: %s",
			      (void *)dev, strerror(ret));
			goto error;
		}
	}
	attr.cq = (struct ibv_exp_cq_init_attr){
		.comp_mask = 0,
	};
	if (priv->cqe_comp) {
		attr.cq.comp_mask |= IBV_EXP_CQ_INIT_ATTR_FLAGS;
		attr.cq.flags |= IBV_EXP_CQ_COMPRESSED_CQE;
		/*
		 * For vectorized Rx, it must not be doubled in order to
		 * make cq_ci and rq_ci aligned.
		 */
		if (rxq_check_vec_support(&tmpl.rxq) < 0)
			cqe_n = (desc * 2) - 1; /* Double the number of CQEs. */
	}
	tmpl.cq = ibv_exp_create_cq(priv->ctx, cqe_n, NULL, tmpl.channel, 0,
				    &attr.cq);
	if (tmpl.cq == NULL) {
		ret = ENOMEM;
		ERROR("%p: CQ creation failure: %s",
		      (void *)dev, strerror(ret));
		goto error;
	}
	DEBUG("priv->device_attr.max_qp_wr is %d",
	      priv->device_attr.max_qp_wr);
	DEBUG("priv->device_attr.max_sge is %d",
	      priv->device_attr.max_sge);
	/* Configure VLAN stripping. */
	tmpl.rxq.vlan_strip = (priv->hw_vlan_strip &&
			       !!dev->data->dev_conf.rxmode.hw_vlan_strip);
	attr.wq = (struct ibv_exp_wq_init_attr){
		.wq_context = NULL, /* Could be useful in the future. */
		.wq_type = IBV_EXP_WQT_RQ,
		/* Max number of outstanding WRs. */
		.max_recv_wr = desc >> tmpl.rxq.sges_n,
		/* Max number of scatter/gather elements in a WR. */
		.max_recv_sge = 1 << tmpl.rxq.sges_n,
		.pd = priv->pd,
		.cq = tmpl.cq,
		.comp_mask =
			IBV_EXP_CREATE_WQ_VLAN_OFFLOADS |
			0,
		.vlan_offloads = (tmpl.rxq.vlan_strip ?
				  IBV_EXP_RECEIVE_WQ_CVLAN_STRIP :
				  0),
	};
	/* By default, FCS (CRC) is stripped by hardware. */
	if (dev->data->dev_conf.rxmode.hw_strip_crc) {
		tmpl.rxq.crc_present = 0;
	} else if (priv->hw_fcs_strip) {
		/* Ask HW/Verbs to leave CRC in place when supported. */
		attr.wq.flags |= IBV_EXP_CREATE_WQ_FLAG_SCATTER_FCS;
		attr.wq.comp_mask |= IBV_EXP_CREATE_WQ_FLAGS;
		tmpl.rxq.crc_present = 1;
	} else {
		WARN("%p: CRC stripping has been disabled but will still"
		     " be performed by hardware, make sure MLNX_OFED and"
		     " firmware are up to date",
		     (void *)dev);
		tmpl.rxq.crc_present = 0;
	}
	DEBUG("%p: CRC stripping is %s, %u bytes will be subtracted from"
	      " incoming frames to hide it",
	      (void *)dev,
	      tmpl.rxq.crc_present ? "disabled" : "enabled",
	      tmpl.rxq.crc_present << 2);
	if (!mlx5_getenv_int("MLX5_PMD_ENABLE_PADDING"))
		; /* Nothing else to do. */
	else if (priv->hw_padding) {
		INFO("%p: enabling packet padding on queue %p",
		     (void *)dev, (void *)rxq_ctrl);
		attr.wq.flags |= IBV_EXP_CREATE_WQ_FLAG_RX_END_PADDING;
		attr.wq.comp_mask |= IBV_EXP_CREATE_WQ_FLAGS;
	} else
		WARN("%p: packet padding has been requested but is not"
		     " supported, make sure MLNX_OFED and firmware are"
		     " up to date",
		     (void *)dev);

	tmpl.wq = ibv_exp_create_wq(priv->ctx, &attr.wq);
	if (tmpl.wq == NULL) {
		ret = (errno ? errno : EINVAL);
		ERROR("%p: WQ creation failure: %s",
		      (void *)dev, strerror(ret));
		goto error;
	}
	/*
	 * Make sure number of WRs*SGEs match expectations since a queue
	 * cannot allocate more than "desc" buffers.
	 */
	if (((int)attr.wq.max_recv_wr != (desc >> tmpl.rxq.sges_n)) ||
	    ((int)attr.wq.max_recv_sge != (1 << tmpl.rxq.sges_n))) {
		ERROR("%p: requested %u*%u but got %u*%u WRs*SGEs",
		      (void *)dev,
		      (desc >> tmpl.rxq.sges_n), (1 << tmpl.rxq.sges_n),
		      attr.wq.max_recv_wr, attr.wq.max_recv_sge);
		ret = EINVAL;
		goto error;
	}
	/* Save port ID. */
	tmpl.rxq.port_id = dev->data->port_id;
	DEBUG("%p: RTE port ID: %u", (void *)rxq_ctrl, tmpl.rxq.port_id);
	/* Change queue state to ready. */
	mod = (struct ibv_exp_wq_attr){
		.attr_mask = IBV_EXP_WQ_ATTR_STATE,
		.wq_state = IBV_EXP_WQS_RDY,
	};
	ret = ibv_exp_modify_wq(tmpl.wq, &mod);
	if (ret) {
		ERROR("%p: WQ state to IBV_EXP_WQS_RDY failed: %s",
		      (void *)dev, strerror(ret));
		goto error;
	}
	ret = rxq_setup(&tmpl);
	if (ret) {
		ERROR("%p: cannot initialize RX queue structure: %s",
		      (void *)dev, strerror(ret));
		goto error;
	}
	ret = rxq_alloc_elts(&tmpl, desc);
	if (ret) {
		ERROR("%p: RXQ allocation failed: %s",
		      (void *)dev, strerror(ret));
		goto error;
	}
	/* Clean up rxq in case we're reinitializing it. */
	DEBUG("%p: cleaning-up old rxq just in case", (void *)rxq_ctrl);
	rxq_cleanup(rxq_ctrl);
	/* Move mbuf pointers to dedicated storage area in RX queue. */
	elts = (void *)(rxq_ctrl + 1);
	rte_memcpy(elts, tmpl.rxq.elts, sizeof(*elts));
#ifndef NDEBUG
	memset(tmpl.rxq.elts, 0x55, sizeof(*elts));
#endif
	rte_free(tmpl.rxq.elts);
	tmpl.rxq.elts = elts;
	*rxq_ctrl = tmpl;
	/* Update doorbell counter. */
	rxq_ctrl->rxq.rq_ci = desc >> rxq_ctrl->rxq.sges_n;
	rte_wmb();
	*rxq_ctrl->rxq.rq_db = htonl(rxq_ctrl->rxq.rq_ci);
	DEBUG("%p: rxq updated with %p", (void *)rxq_ctrl, (void *)&tmpl);
	assert(ret == 0);
	return 0;
error:
	elts = tmpl.rxq.elts;
	rxq_cleanup(&tmpl);
	rte_free(elts);
	assert(ret > 0);
	return ret;
}

/**
 * DPDK callback to configure a RX queue.
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 * @param idx
 *   RX queue index.
 * @param desc
 *   Number of descriptors to configure in queue.
 * @param socket
 *   NUMA socket on which memory must be allocated.
 * @param[in] conf
 *   Thresholds parameters.
 * @param mp
 *   Memory pool for buffer allocations.
 *
 * @return
 *   0 on success, negative errno value on failure.
 */
int
mlx5_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
		    unsigned int socket, const struct rte_eth_rxconf *conf,
		    struct rte_mempool *mp)
{
	struct priv *priv = dev->data->dev_private;
	struct rxq *rxq = (*priv->rxqs)[idx];
	struct rxq_ctrl *rxq_ctrl = container_of(rxq, struct rxq_ctrl, rxq);
	const uint16_t desc_n =
		desc + priv->rx_vec_en * MLX5_VPMD_DESCS_PER_LOOP;
	int ret;

	if (mlx5_is_secondary())
		return -E_RTE_SECONDARY;

	priv_lock(priv);
	if (!rte_is_power_of_2(desc)) {
		desc = 1 << log2above(desc);
		WARN("%p: increased number of descriptors in RX queue %u"
		     " to the next power of two (%d)",
		     (void *)dev, idx, desc);
	}
	DEBUG("%p: configuring queue %u for %u descriptors",
	      (void *)dev, idx, desc);
	if (idx >= priv->rxqs_n) {
		ERROR("%p: queue index out of range (%u >= %u)",
		      (void *)dev, idx, priv->rxqs_n);
		priv_unlock(priv);
		return -EOVERFLOW;
	}
	if (rxq != NULL) {
		DEBUG("%p: reusing already allocated queue index %u (%p)",
		      (void *)dev, idx, (void *)rxq);
		if (priv->started) {
			priv_unlock(priv);
			return -EEXIST;
		}
		(*priv->rxqs)[idx] = NULL;
		rxq_cleanup(rxq_ctrl);
		/* Resize if rxq size is changed. */
		if (rxq_ctrl->rxq.elts_n != log2above(desc)) {
			rxq_ctrl = rte_realloc(rxq_ctrl,
					       sizeof(*rxq_ctrl) + desc_n *
					       sizeof(struct rte_mbuf *),
					       RTE_CACHE_LINE_SIZE);
			if (!rxq_ctrl) {
				ERROR("%p: unable to reallocate queue index %u",
					(void *)dev, idx);
				priv_unlock(priv);
				return -ENOMEM;
			}
		}
	} else {
		rxq_ctrl = rte_calloc_socket("RXQ", 1, sizeof(*rxq_ctrl) +
					     desc_n *
					     sizeof(struct rte_mbuf *),
					     0, socket);
		if (rxq_ctrl == NULL) {
			ERROR("%p: unable to allocate queue index %u",
			      (void *)dev, idx);
			priv_unlock(priv);
			return -ENOMEM;
		}
	}
	ret = rxq_ctrl_setup(dev, rxq_ctrl, desc, socket, conf, mp);
	if (ret)
		rte_free(rxq_ctrl);
	else {
		rxq_ctrl->rxq.stats.idx = idx;
		DEBUG("%p: adding RX queue %p to list",
		      (void *)dev, (void *)rxq_ctrl);
		(*priv->rxqs)[idx] = &rxq_ctrl->rxq;
	}
	priv_unlock(priv);
	return -ret;
}

/**
 * DPDK callback to release a RX queue.
 *
 * @param dpdk_rxq
 *   Generic RX queue pointer.
 */
void
mlx5_rx_queue_release(void *dpdk_rxq)
{
	struct rxq *rxq = (struct rxq *)dpdk_rxq;
	struct rxq_ctrl *rxq_ctrl;
	struct priv *priv;
	unsigned int i;

	if (mlx5_is_secondary())
		return;

	if (rxq == NULL)
		return;
	rxq_ctrl = container_of(rxq, struct rxq_ctrl, rxq);
	priv = rxq_ctrl->priv;
	priv_lock(priv);
	if (priv_flow_rxq_in_use(priv, rxq))
		rte_panic("Rx queue %p is still used by a flow and cannot be"
			  " removed\n", (void *)rxq_ctrl);
	for (i = 0; (i != priv->rxqs_n); ++i)
		if ((*priv->rxqs)[i] == rxq) {
			DEBUG("%p: removing RX queue %p from list",
			      (void *)priv->dev, (void *)rxq_ctrl);
			(*priv->rxqs)[i] = NULL;
			break;
		}
	rxq_cleanup(rxq_ctrl);
	rte_free(rxq_ctrl);
	priv_unlock(priv);
}

/**
 * DPDK callback for RX in secondary processes.
 *
 * This function configures all queues from primary process information
 * if necessary before reverting to the normal RX burst callback.
 *
 * @param dpdk_rxq
 *   Generic pointer to RX queue structure.
 * @param[out] pkts
 *   Array to store received packets.
 * @param pkts_n
 *   Maximum number of packets in array.
 *
 * @return
 *   Number of packets successfully received (<= pkts_n).
 */
uint16_t
mlx5_rx_burst_secondary_setup(void *dpdk_rxq, struct rte_mbuf **pkts,
			      uint16_t pkts_n)
{
	struct rxq *rxq = dpdk_rxq;
	struct rxq_ctrl *rxq_ctrl = container_of(rxq, struct rxq_ctrl, rxq);
	struct priv *priv = mlx5_secondary_data_setup(rxq_ctrl->priv);
	struct priv *primary_priv;
	unsigned int index;

	if (priv == NULL)
		return 0;
	primary_priv =
		mlx5_secondary_data[priv->dev->data->port_id].primary_priv;
	/* Look for queue index in both private structures. */
	for (index = 0; index != priv->rxqs_n; ++index)
		if (((*primary_priv->rxqs)[index] == rxq) ||
		    ((*priv->rxqs)[index] == rxq))
			break;
	if (index == priv->rxqs_n)
		return 0;
	rxq = (*priv->rxqs)[index];
	return priv->dev->rx_pkt_burst(rxq, pkts, pkts_n);
}

/**
 * Allocate queue vector and fill epoll fd list for Rx interrupts.
 *
 * @param priv
 *   Pointer to private structure.
 *
 * @return
 *   0 on success, negative on failure.
 */
int
priv_rx_intr_vec_enable(struct priv *priv)
{
	unsigned int i;
	unsigned int rxqs_n = priv->rxqs_n;
	unsigned int n = RTE_MIN(rxqs_n, (uint32_t)RTE_MAX_RXTX_INTR_VEC_ID);
	unsigned int count = 0;
	struct rte_intr_handle *intr_handle = priv->dev->intr_handle;

	if (!priv->dev->data->dev_conf.intr_conf.rxq)
		return 0;
	priv_rx_intr_vec_disable(priv);
	intr_handle->intr_vec = malloc(sizeof(intr_handle->intr_vec[rxqs_n]));
	if (intr_handle->intr_vec == NULL) {
		ERROR("failed to allocate memory for interrupt vector,"
		      " Rx interrupts will not be supported");
		return -ENOMEM;
	}
	intr_handle->type = RTE_INTR_HANDLE_EXT;
	for (i = 0; i != n; ++i) {
		struct rxq *rxq = (*priv->rxqs)[i];
		struct rxq_ctrl *rxq_ctrl =
			container_of(rxq, struct rxq_ctrl, rxq);
		int fd;
		int flags;
		int rc;

		/* Skip queues that cannot request interrupts. */
		if (!rxq || !rxq_ctrl->channel) {
			/* Use invalid intr_vec[] index to disable entry. */
			intr_handle->intr_vec[i] =
				RTE_INTR_VEC_RXTX_OFFSET +
				RTE_MAX_RXTX_INTR_VEC_ID;
			continue;
		}
		if (count >= RTE_MAX_RXTX_INTR_VEC_ID) {
			ERROR("too many Rx queues for interrupt vector size"
			      " (%d), Rx interrupts cannot be enabled",
			      RTE_MAX_RXTX_INTR_VEC_ID);
			priv_rx_intr_vec_disable(priv);
			return -1;
		}
		fd = rxq_ctrl->channel->fd;
		flags = fcntl(fd, F_GETFL);
		rc = fcntl(fd, F_SETFL, flags | O_NONBLOCK);
		if (rc < 0) {
			ERROR("failed to make Rx interrupt file descriptor"
			      " %d non-blocking for queue index %d", fd, i);
			priv_rx_intr_vec_disable(priv);
			return -1;
		}
		intr_handle->intr_vec[i] = RTE_INTR_VEC_RXTX_OFFSET + count;
		intr_handle->efds[count] = fd;
		count++;
	}
	if (!count)
		priv_rx_intr_vec_disable(priv);
	else
		intr_handle->nb_efd = count;
	return 0;
}

/**
 * Clean up Rx interrupts handler.
 *
 * @param priv
 *   Pointer to private structure.
 */
void
priv_rx_intr_vec_disable(struct priv *priv)
{
	struct rte_intr_handle *intr_handle = priv->dev->intr_handle;

	rte_intr_free_epoll_fd(intr_handle);
	free(intr_handle->intr_vec);
	intr_handle->nb_efd = 0;
	intr_handle->intr_vec = NULL;
}

#ifdef HAVE_UPDATE_CQ_CI

/**
 * DPDK callback for Rx queue interrupt enable.
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 * @param rx_queue_id
 *   Rx queue number.
 *
 * @return
 *   0 on success, negative on failure.
 */
int
mlx5_rx_intr_enable(struct rte_eth_dev *dev, uint16_t rx_queue_id)
{
	struct priv *priv = mlx5_get_priv(dev);
	struct rxq *rxq = (*priv->rxqs)[rx_queue_id];
	struct rxq_ctrl *rxq_ctrl = container_of(rxq, struct rxq_ctrl, rxq);
	int ret;

	if (!rxq || !rxq_ctrl->channel) {
		ret = EINVAL;
	} else {
		ibv_mlx5_exp_update_cq_ci(rxq_ctrl->cq, rxq->cq_ci);
		ret = ibv_req_notify_cq(rxq_ctrl->cq, 0);
	}
	if (ret)
		WARN("unable to arm interrupt on rx queue %d", rx_queue_id);
	return -ret;
}

/**
 * DPDK callback for Rx queue interrupt disable.
 *
 * @param dev
 *   Pointer to Ethernet device structure.
 * @param rx_queue_id
 *   Rx queue number.
 *
 * @return
 *   0 on success, negative on failure.
 */
int
mlx5_rx_intr_disable(struct rte_eth_dev *dev, uint16_t rx_queue_id)
{
	struct priv *priv = mlx5_get_priv(dev);
	struct rxq *rxq = (*priv->rxqs)[rx_queue_id];
	struct rxq_ctrl *rxq_ctrl = container_of(rxq, struct rxq_ctrl, rxq);
	struct ibv_cq *ev_cq;
	void *ev_ctx;
	int ret;

	if (!rxq || !rxq_ctrl->channel) {
		ret = EINVAL;
	} else {
		ret = ibv_get_cq_event(rxq_ctrl->cq->channel, &ev_cq, &ev_ctx);
		if (ret || ev_cq != rxq_ctrl->cq)
			ret = EINVAL;
	}
	if (ret)
		WARN("unable to disable interrupt on rx queue %d",
		     rx_queue_id);
	else
		ibv_ack_cq_events(rxq_ctrl->cq, 1);
	return -ret;
}

#endif /* HAVE_UPDATE_CQ_CI */