summaryrefslogtreecommitdiff
path: root/drivers/net/sfc/sfc_tx.c
blob: 1cd08d8c4138a6ee11cf1375e66a6f15f098d747 (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
/* SPDX-License-Identifier: BSD-3-Clause
 *
 * Copyright (c) 2016-2018 Solarflare Communications Inc.
 * All rights reserved.
 *
 * This software was jointly developed between OKTET Labs (under contract
 * for Solarflare) and Solarflare Communications, Inc.
 */

#include "sfc.h"
#include "sfc_debug.h"
#include "sfc_log.h"
#include "sfc_ev.h"
#include "sfc_tx.h"
#include "sfc_tweak.h"
#include "sfc_kvargs.h"

/*
 * Maximum number of TX queue flush attempts in case of
 * failure or flush timeout
 */
#define SFC_TX_QFLUSH_ATTEMPTS		(3)

/*
 * Time to wait between event queue polling attempts when waiting for TX
 * queue flush done or flush failed events
 */
#define SFC_TX_QFLUSH_POLL_WAIT_MS	(1)

/*
 * Maximum number of event queue polling attempts when waiting for TX queue
 * flush done or flush failed events; it defines TX queue flush attempt timeout
 * together with SFC_TX_QFLUSH_POLL_WAIT_MS
 */
#define SFC_TX_QFLUSH_POLL_ATTEMPTS	(2000)

uint64_t
sfc_tx_get_dev_offload_caps(struct sfc_adapter *sa)
{
	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
	uint64_t caps = 0;

	if ((sa->dp_tx->features & SFC_DP_TX_FEAT_VLAN_INSERT) &&
	    encp->enc_hw_tx_insert_vlan_enabled)
		caps |= DEV_TX_OFFLOAD_VLAN_INSERT;

	if (sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_SEG)
		caps |= DEV_TX_OFFLOAD_MULTI_SEGS;

	if ((~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_POOL) &&
	    (~sa->dp_tx->features & SFC_DP_TX_FEAT_REFCNT))
		caps |= DEV_TX_OFFLOAD_MBUF_FAST_FREE;

	return caps;
}

uint64_t
sfc_tx_get_queue_offload_caps(struct sfc_adapter *sa)
{
	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
	uint64_t caps = 0;

	caps |= DEV_TX_OFFLOAD_IPV4_CKSUM;
	caps |= DEV_TX_OFFLOAD_UDP_CKSUM;
	caps |= DEV_TX_OFFLOAD_TCP_CKSUM;

	if (encp->enc_tunnel_encapsulations_supported)
		caps |= DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM;

	if (sa->tso)
		caps |= DEV_TX_OFFLOAD_TCP_TSO;

	return caps;
}

static void
sfc_tx_log_offloads(struct sfc_adapter *sa, const char *offload_group,
		    const char *verdict, uint64_t offloads)
{
	unsigned long long bit;

	while ((bit = __builtin_ffsll(offloads)) != 0) {
		uint64_t flag = (1ULL << --bit);

		sfc_err(sa, "Tx %s offload %s %s", offload_group,
			rte_eth_dev_tx_offload_name(flag), verdict);

		offloads &= ~flag;
	}
}

static int
sfc_tx_queue_offload_mismatch(struct sfc_adapter *sa, uint64_t requested)
{
	uint64_t mandatory = sa->eth_dev->data->dev_conf.txmode.offloads;
	uint64_t supported = sfc_tx_get_dev_offload_caps(sa) |
			     sfc_tx_get_queue_offload_caps(sa);
	uint64_t rejected = requested & ~supported;
	uint64_t missing = (requested & mandatory) ^ mandatory;
	boolean_t mismatch = B_FALSE;

	if (rejected) {
		sfc_tx_log_offloads(sa, "queue", "is unsupported", rejected);
		mismatch = B_TRUE;
	}

	if (missing) {
		sfc_tx_log_offloads(sa, "queue", "must be set", missing);
		mismatch = B_TRUE;
	}

	return mismatch;
}

static int
sfc_tx_qcheck_conf(struct sfc_adapter *sa, unsigned int txq_max_fill_level,
		   const struct rte_eth_txconf *tx_conf)
{
	int rc = 0;

	if (tx_conf->tx_rs_thresh != 0) {
		sfc_err(sa, "RS bit in transmit descriptor is not supported");
		rc = EINVAL;
	}

	if (tx_conf->tx_free_thresh > txq_max_fill_level) {
		sfc_err(sa,
			"TxQ free threshold too large: %u vs maximum %u",
			tx_conf->tx_free_thresh, txq_max_fill_level);
		rc = EINVAL;
	}

	if (tx_conf->tx_thresh.pthresh != 0 ||
	    tx_conf->tx_thresh.hthresh != 0 ||
	    tx_conf->tx_thresh.wthresh != 0) {
		sfc_warn(sa,
			"prefetch/host/writeback thresholds are not supported");
	}

	/* We either perform both TCP and UDP offload, or no offload at all */
	if (((tx_conf->offloads & DEV_TX_OFFLOAD_TCP_CKSUM) == 0) !=
	    ((tx_conf->offloads & DEV_TX_OFFLOAD_UDP_CKSUM) == 0)) {
		sfc_err(sa, "TCP and UDP offloads can't be set independently");
		rc = EINVAL;
	}

	if (sfc_tx_queue_offload_mismatch(sa, tx_conf->offloads))
		rc = EINVAL;

	return rc;
}

void
sfc_tx_qflush_done(struct sfc_txq *txq)
{
	txq->state |= SFC_TXQ_FLUSHED;
	txq->state &= ~SFC_TXQ_FLUSHING;
}

int
sfc_tx_qinit(struct sfc_adapter *sa, unsigned int sw_index,
	     uint16_t nb_tx_desc, unsigned int socket_id,
	     const struct rte_eth_txconf *tx_conf)
{
	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
	unsigned int txq_entries;
	unsigned int evq_entries;
	unsigned int txq_max_fill_level;
	struct sfc_txq_info *txq_info;
	struct sfc_evq *evq;
	struct sfc_txq *txq;
	int rc = 0;
	struct sfc_dp_tx_qcreate_info info;

	sfc_log_init(sa, "TxQ = %u", sw_index);

	rc = sa->dp_tx->qsize_up_rings(nb_tx_desc, &txq_entries, &evq_entries,
				       &txq_max_fill_level);
	if (rc != 0)
		goto fail_size_up_rings;
	SFC_ASSERT(txq_entries >= EFX_TXQ_MINNDESCS);
	SFC_ASSERT(txq_entries <= sa->txq_max_entries);
	SFC_ASSERT(txq_entries >= nb_tx_desc);
	SFC_ASSERT(txq_max_fill_level <= nb_tx_desc);

	rc = sfc_tx_qcheck_conf(sa, txq_max_fill_level, tx_conf);
	if (rc != 0)
		goto fail_bad_conf;

	SFC_ASSERT(sw_index < sa->txq_count);
	txq_info = &sa->txq_info[sw_index];

	txq_info->entries = txq_entries;

	rc = sfc_ev_qinit(sa, SFC_EVQ_TYPE_TX, sw_index,
			  evq_entries, socket_id, &evq);
	if (rc != 0)
		goto fail_ev_qinit;

	rc = ENOMEM;
	txq = rte_zmalloc_socket("sfc-txq", sizeof(*txq), 0, socket_id);
	if (txq == NULL)
		goto fail_txq_alloc;

	txq_info->txq = txq;

	txq->hw_index = sw_index;
	txq->evq = evq;
	txq->free_thresh =
		(tx_conf->tx_free_thresh) ? tx_conf->tx_free_thresh :
		SFC_TX_DEFAULT_FREE_THRESH;
	txq->flags = tx_conf->txq_flags;
	txq->offloads = tx_conf->offloads;

	rc = sfc_dma_alloc(sa, "txq", sw_index, EFX_TXQ_SIZE(txq_info->entries),
			   socket_id, &txq->mem);
	if (rc != 0)
		goto fail_dma_alloc;

	memset(&info, 0, sizeof(info));
	info.max_fill_level = txq_max_fill_level;
	info.free_thresh = txq->free_thresh;
	info.flags = tx_conf->txq_flags;
	info.offloads = tx_conf->offloads;
	info.txq_entries = txq_info->entries;
	info.dma_desc_size_max = encp->enc_tx_dma_desc_size_max;
	info.txq_hw_ring = txq->mem.esm_base;
	info.evq_entries = evq_entries;
	info.evq_hw_ring = evq->mem.esm_base;
	info.hw_index = txq->hw_index;
	info.mem_bar = sa->mem_bar.esb_base;
	info.vi_window_shift = encp->enc_vi_window_shift;

	rc = sa->dp_tx->qcreate(sa->eth_dev->data->port_id, sw_index,
				&RTE_ETH_DEV_TO_PCI(sa->eth_dev)->addr,
				socket_id, &info, &txq->dp);
	if (rc != 0)
		goto fail_dp_tx_qinit;

	evq->dp_txq = txq->dp;

	txq->state = SFC_TXQ_INITIALIZED;

	txq_info->deferred_start = (tx_conf->tx_deferred_start != 0);

	return 0;

fail_dp_tx_qinit:
	sfc_dma_free(sa, &txq->mem);

fail_dma_alloc:
	txq_info->txq = NULL;
	rte_free(txq);

fail_txq_alloc:
	sfc_ev_qfini(evq);

fail_ev_qinit:
	txq_info->entries = 0;

fail_bad_conf:
fail_size_up_rings:
	sfc_log_init(sa, "failed (TxQ = %u, rc = %d)", sw_index, rc);
	return rc;
}

void
sfc_tx_qfini(struct sfc_adapter *sa, unsigned int sw_index)
{
	struct sfc_txq_info *txq_info;
	struct sfc_txq *txq;

	sfc_log_init(sa, "TxQ = %u", sw_index);

	SFC_ASSERT(sw_index < sa->txq_count);
	txq_info = &sa->txq_info[sw_index];

	txq = txq_info->txq;
	SFC_ASSERT(txq != NULL);
	SFC_ASSERT(txq->state == SFC_TXQ_INITIALIZED);

	sa->dp_tx->qdestroy(txq->dp);
	txq->dp = NULL;

	txq_info->txq = NULL;
	txq_info->entries = 0;

	sfc_dma_free(sa, &txq->mem);

	sfc_ev_qfini(txq->evq);
	txq->evq = NULL;

	rte_free(txq);
}

static int
sfc_tx_qinit_info(struct sfc_adapter *sa, unsigned int sw_index)
{
	sfc_log_init(sa, "TxQ = %u", sw_index);

	return 0;
}

static int
sfc_tx_check_mode(struct sfc_adapter *sa, const struct rte_eth_txmode *txmode)
{
	uint64_t offloads_supported = sfc_tx_get_dev_offload_caps(sa) |
				      sfc_tx_get_queue_offload_caps(sa);
	uint64_t offloads_rejected = txmode->offloads & ~offloads_supported;
	int rc = 0;

	switch (txmode->mq_mode) {
	case ETH_MQ_TX_NONE:
		break;
	default:
		sfc_err(sa, "Tx multi-queue mode %u not supported",
			txmode->mq_mode);
		rc = EINVAL;
	}

	/*
	 * These features are claimed to be i40e-specific,
	 * but it does make sense to double-check their absence
	 */
	if (txmode->hw_vlan_reject_tagged) {
		sfc_err(sa, "Rejecting tagged packets not supported");
		rc = EINVAL;
	}

	if (txmode->hw_vlan_reject_untagged) {
		sfc_err(sa, "Rejecting untagged packets not supported");
		rc = EINVAL;
	}

	if (txmode->hw_vlan_insert_pvid) {
		sfc_err(sa, "Port-based VLAN insertion not supported");
		rc = EINVAL;
	}

	if (offloads_rejected) {
		sfc_tx_log_offloads(sa, "device", "is unsupported",
				    offloads_rejected);
		rc = EINVAL;
	}

	return rc;
}

/**
 * Destroy excess queues that are no longer needed after reconfiguration
 * or complete close.
 */
static void
sfc_tx_fini_queues(struct sfc_adapter *sa, unsigned int nb_tx_queues)
{
	int sw_index;

	SFC_ASSERT(nb_tx_queues <= sa->txq_count);

	sw_index = sa->txq_count;
	while (--sw_index >= (int)nb_tx_queues) {
		if (sa->txq_info[sw_index].txq != NULL)
			sfc_tx_qfini(sa, sw_index);
	}

	sa->txq_count = nb_tx_queues;
}

int
sfc_tx_configure(struct sfc_adapter *sa)
{
	const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
	const struct rte_eth_conf *dev_conf = &sa->eth_dev->data->dev_conf;
	const unsigned int nb_tx_queues = sa->eth_dev->data->nb_tx_queues;
	int rc = 0;

	sfc_log_init(sa, "nb_tx_queues=%u (old %u)",
		     nb_tx_queues, sa->txq_count);

	/*
	 * The datapath implementation assumes absence of boundary
	 * limits on Tx DMA descriptors. Addition of these checks on
	 * datapath would simply make the datapath slower.
	 */
	if (encp->enc_tx_dma_desc_boundary != 0) {
		rc = ENOTSUP;
		goto fail_tx_dma_desc_boundary;
	}

	rc = sfc_tx_check_mode(sa, &dev_conf->txmode);
	if (rc != 0)
		goto fail_check_mode;

	if (nb_tx_queues == sa->txq_count)
		goto done;

	if (sa->txq_info == NULL) {
		sa->txq_info = rte_calloc_socket("sfc-txqs", nb_tx_queues,
						 sizeof(sa->txq_info[0]), 0,
						 sa->socket_id);
		if (sa->txq_info == NULL)
			goto fail_txqs_alloc;
	} else {
		struct sfc_txq_info *new_txq_info;

		if (nb_tx_queues < sa->txq_count)
			sfc_tx_fini_queues(sa, nb_tx_queues);

		new_txq_info =
			rte_realloc(sa->txq_info,
				    nb_tx_queues * sizeof(sa->txq_info[0]), 0);
		if (new_txq_info == NULL && nb_tx_queues > 0)
			goto fail_txqs_realloc;

		sa->txq_info = new_txq_info;
		if (nb_tx_queues > sa->txq_count)
			memset(&sa->txq_info[sa->txq_count], 0,
			       (nb_tx_queues - sa->txq_count) *
			       sizeof(sa->txq_info[0]));
	}

	while (sa->txq_count < nb_tx_queues) {
		rc = sfc_tx_qinit_info(sa, sa->txq_count);
		if (rc != 0)
			goto fail_tx_qinit_info;

		sa->txq_count++;
	}

done:
	return 0;

fail_tx_qinit_info:
fail_txqs_realloc:
fail_txqs_alloc:
	sfc_tx_close(sa);

fail_check_mode:
fail_tx_dma_desc_boundary:
	sfc_log_init(sa, "failed (rc = %d)", rc);
	return rc;
}

void
sfc_tx_close(struct sfc_adapter *sa)
{
	sfc_tx_fini_queues(sa, 0);

	rte_free(sa->txq_info);
	sa->txq_info = NULL;
}

int
sfc_tx_qstart(struct sfc_adapter *sa, unsigned int sw_index)
{
	uint64_t offloads_supported = sfc_tx_get_dev_offload_caps(sa) |
				      sfc_tx_get_queue_offload_caps(sa);
	struct rte_eth_dev_data *dev_data;
	struct sfc_txq_info *txq_info;
	struct sfc_txq *txq;
	struct sfc_evq *evq;
	uint16_t flags = 0;
	unsigned int desc_index;
	int rc = 0;

	sfc_log_init(sa, "TxQ = %u", sw_index);

	SFC_ASSERT(sw_index < sa->txq_count);
	txq_info = &sa->txq_info[sw_index];

	txq = txq_info->txq;

	SFC_ASSERT(txq->state == SFC_TXQ_INITIALIZED);

	evq = txq->evq;

	rc = sfc_ev_qstart(evq, sfc_evq_index_by_txq_sw_index(sa, sw_index));
	if (rc != 0)
		goto fail_ev_qstart;

	/*
	 * The absence of ETH_TXQ_FLAGS_IGNORE is associated with a legacy
	 * application which expects that IPv4 checksum offload is enabled
	 * all the time as there is no legacy flag to turn off the offload.
	 */
	if ((txq->offloads & DEV_TX_OFFLOAD_IPV4_CKSUM) ||
	    (~txq->flags & ETH_TXQ_FLAGS_IGNORE))
		flags |= EFX_TXQ_CKSUM_IPV4;

	if ((txq->offloads & DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM) ||
	    ((~txq->flags & ETH_TXQ_FLAGS_IGNORE) &&
	     (offloads_supported & DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM)))
		flags |= EFX_TXQ_CKSUM_INNER_IPV4;

	if ((txq->offloads & DEV_TX_OFFLOAD_TCP_CKSUM) ||
	    (txq->offloads & DEV_TX_OFFLOAD_UDP_CKSUM)) {
		flags |= EFX_TXQ_CKSUM_TCPUDP;

		if ((~txq->flags & ETH_TXQ_FLAGS_IGNORE) &&
		    (offloads_supported & DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM))
			flags |= EFX_TXQ_CKSUM_INNER_TCPUDP;
	}

	/*
	 * The absence of ETH_TXQ_FLAGS_IGNORE is associated with a legacy
	 * application. In turn, the absence of ETH_TXQ_FLAGS_NOXSUMTCP is
	 * associated specifically with a legacy application which expects
	 * both TCP checksum offload and TSO to be enabled because the legacy
	 * API does not provide a dedicated mechanism to control TSO.
	 */
	if ((txq->offloads & DEV_TX_OFFLOAD_TCP_TSO) ||
	    ((~txq->flags & ETH_TXQ_FLAGS_IGNORE) &&
	     (~txq->flags & ETH_TXQ_FLAGS_NOXSUMTCP)))
		flags |= EFX_TXQ_FATSOV2;

	rc = efx_tx_qcreate(sa->nic, sw_index, 0, &txq->mem,
			    txq_info->entries, 0 /* not used on EF10 */,
			    flags, evq->common,
			    &txq->common, &desc_index);
	if (rc != 0) {
		if (sa->tso && (rc == ENOSPC))
			sfc_err(sa, "ran out of TSO contexts");

		goto fail_tx_qcreate;
	}

	efx_tx_qenable(txq->common);

	txq->state |= SFC_TXQ_STARTED;

	rc = sa->dp_tx->qstart(txq->dp, evq->read_ptr, desc_index);
	if (rc != 0)
		goto fail_dp_qstart;

	/*
	 * It seems to be used by DPDK for debug purposes only ('rte_ether')
	 */
	dev_data = sa->eth_dev->data;
	dev_data->tx_queue_state[sw_index] = RTE_ETH_QUEUE_STATE_STARTED;

	return 0;

fail_dp_qstart:
	txq->state = SFC_TXQ_INITIALIZED;
	efx_tx_qdestroy(txq->common);

fail_tx_qcreate:
	sfc_ev_qstop(evq);

fail_ev_qstart:
	return rc;
}

void
sfc_tx_qstop(struct sfc_adapter *sa, unsigned int sw_index)
{
	struct rte_eth_dev_data *dev_data;
	struct sfc_txq_info *txq_info;
	struct sfc_txq *txq;
	unsigned int retry_count;
	unsigned int wait_count;
	int rc;

	sfc_log_init(sa, "TxQ = %u", sw_index);

	SFC_ASSERT(sw_index < sa->txq_count);
	txq_info = &sa->txq_info[sw_index];

	txq = txq_info->txq;

	if (txq->state == SFC_TXQ_INITIALIZED)
		return;

	SFC_ASSERT(txq->state & SFC_TXQ_STARTED);

	sa->dp_tx->qstop(txq->dp, &txq->evq->read_ptr);

	/*
	 * Retry TX queue flushing in case of flush failed or
	 * timeout; in the worst case it can delay for 6 seconds
	 */
	for (retry_count = 0;
	     ((txq->state & SFC_TXQ_FLUSHED) == 0) &&
	     (retry_count < SFC_TX_QFLUSH_ATTEMPTS);
	     ++retry_count) {
		rc = efx_tx_qflush(txq->common);
		if (rc != 0) {
			txq->state |= (rc == EALREADY) ?
				SFC_TXQ_FLUSHED : SFC_TXQ_FLUSH_FAILED;
			break;
		}

		/*
		 * Wait for TX queue flush done or flush failed event at least
		 * SFC_TX_QFLUSH_POLL_WAIT_MS milliseconds and not more
		 * than 2 seconds (SFC_TX_QFLUSH_POLL_WAIT_MS multiplied
		 * by SFC_TX_QFLUSH_POLL_ATTEMPTS)
		 */
		wait_count = 0;
		do {
			rte_delay_ms(SFC_TX_QFLUSH_POLL_WAIT_MS);
			sfc_ev_qpoll(txq->evq);
		} while ((txq->state & SFC_TXQ_FLUSHING) &&
			 wait_count++ < SFC_TX_QFLUSH_POLL_ATTEMPTS);

		if (txq->state & SFC_TXQ_FLUSHING)
			sfc_err(sa, "TxQ %u flush timed out", sw_index);

		if (txq->state & SFC_TXQ_FLUSHED)
			sfc_notice(sa, "TxQ %u flushed", sw_index);
	}

	sa->dp_tx->qreap(txq->dp);

	txq->state = SFC_TXQ_INITIALIZED;

	efx_tx_qdestroy(txq->common);

	sfc_ev_qstop(txq->evq);

	/*
	 * It seems to be used by DPDK for debug purposes only ('rte_ether')
	 */
	dev_data = sa->eth_dev->data;
	dev_data->tx_queue_state[sw_index] = RTE_ETH_QUEUE_STATE_STOPPED;
}

int
sfc_tx_start(struct sfc_adapter *sa)
{
	unsigned int sw_index;
	int rc = 0;

	sfc_log_init(sa, "txq_count = %u", sa->txq_count);

	if (sa->tso) {
		if (!efx_nic_cfg_get(sa->nic)->enc_fw_assisted_tso_v2_enabled) {
			sfc_warn(sa, "TSO support was unable to be restored");
			sa->tso = B_FALSE;
		}
	}

	rc = efx_tx_init(sa->nic);
	if (rc != 0)
		goto fail_efx_tx_init;

	for (sw_index = 0; sw_index < sa->txq_count; ++sw_index) {
		if (!(sa->txq_info[sw_index].deferred_start) ||
		    sa->txq_info[sw_index].deferred_started) {
			rc = sfc_tx_qstart(sa, sw_index);
			if (rc != 0)
				goto fail_tx_qstart;
		}
	}

	return 0;

fail_tx_qstart:
	while (sw_index-- > 0)
		sfc_tx_qstop(sa, sw_index);

	efx_tx_fini(sa->nic);

fail_efx_tx_init:
	sfc_log_init(sa, "failed (rc = %d)", rc);
	return rc;
}

void
sfc_tx_stop(struct sfc_adapter *sa)
{
	unsigned int sw_index;

	sfc_log_init(sa, "txq_count = %u", sa->txq_count);

	sw_index = sa->txq_count;
	while (sw_index-- > 0) {
		if (sa->txq_info[sw_index].txq != NULL)
			sfc_tx_qstop(sa, sw_index);
	}

	efx_tx_fini(sa->nic);
}

static void
sfc_efx_tx_reap(struct sfc_efx_txq *txq)
{
	unsigned int completed;

	sfc_ev_qpoll(txq->evq);

	for (completed = txq->completed;
	     completed != txq->pending; completed++) {
		struct sfc_efx_tx_sw_desc *txd;

		txd = &txq->sw_ring[completed & txq->ptr_mask];

		if (txd->mbuf != NULL) {
			rte_pktmbuf_free(txd->mbuf);
			txd->mbuf = NULL;
		}
	}

	txq->completed = completed;
}

/*
 * The function is used to insert or update VLAN tag;
 * the firmware has state of the firmware tag to insert per TxQ
 * (controlled by option descriptors), hence, if the tag of the
 * packet to be sent is different from one remembered by the firmware,
 * the function will update it
 */
static unsigned int
sfc_efx_tx_maybe_insert_tag(struct sfc_efx_txq *txq, struct rte_mbuf *m,
			    efx_desc_t **pend)
{
	uint16_t this_tag = ((m->ol_flags & PKT_TX_VLAN_PKT) ?
			     m->vlan_tci : 0);

	if (this_tag == txq->hw_vlan_tci)
		return 0;

	/*
	 * The expression inside SFC_ASSERT() is not desired to be checked in
	 * a non-debug build because it might be too expensive on the data path
	 */
	SFC_ASSERT(efx_nic_cfg_get(txq->evq->sa->nic)->enc_hw_tx_insert_vlan_enabled);

	efx_tx_qdesc_vlantci_create(txq->common, rte_cpu_to_be_16(this_tag),
				    *pend);
	(*pend)++;
	txq->hw_vlan_tci = this_tag;

	return 1;
}

static uint16_t
sfc_efx_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
{
	struct sfc_dp_txq *dp_txq = (struct sfc_dp_txq *)tx_queue;
	struct sfc_efx_txq *txq = sfc_efx_txq_by_dp_txq(dp_txq);
	unsigned int added = txq->added;
	unsigned int pushed = added;
	unsigned int pkts_sent = 0;
	efx_desc_t *pend = &txq->pend_desc[0];
	const unsigned int hard_max_fill = txq->max_fill_level;
	const unsigned int soft_max_fill = hard_max_fill - txq->free_thresh;
	unsigned int fill_level = added - txq->completed;
	boolean_t reap_done;
	int rc __rte_unused;
	struct rte_mbuf **pktp;

	if (unlikely((txq->flags & SFC_EFX_TXQ_FLAG_RUNNING) == 0))
		goto done;

	/*
	 * If insufficient space for a single packet is present,
	 * we should reap; otherwise, we shouldn't do that all the time
	 * to avoid latency increase
	 */
	reap_done = (fill_level > soft_max_fill);

	if (reap_done) {
		sfc_efx_tx_reap(txq);
		/*
		 * Recalculate fill level since 'txq->completed'
		 * might have changed on reap
		 */
		fill_level = added - txq->completed;
	}

	for (pkts_sent = 0, pktp = &tx_pkts[0];
	     (pkts_sent < nb_pkts) && (fill_level <= soft_max_fill);
	     pkts_sent++, pktp++) {
		struct rte_mbuf		*m_seg = *pktp;
		size_t			pkt_len = m_seg->pkt_len;
		unsigned int		pkt_descs = 0;
		size_t			in_off = 0;

		/*
		 * Here VLAN TCI is expected to be zero in case if no
		 * DEV_TX_OFFLOAD_VLAN_INSERT capability is advertised;
		 * if the calling app ignores the absence of
		 * DEV_TX_OFFLOAD_VLAN_INSERT and pushes VLAN TCI, then
		 * TX_ERROR will occur
		 */
		pkt_descs += sfc_efx_tx_maybe_insert_tag(txq, m_seg, &pend);

		if (m_seg->ol_flags & PKT_TX_TCP_SEG) {
			/*
			 * We expect correct 'pkt->l[2, 3, 4]_len' values
			 * to be set correctly by the caller
			 */
			if (sfc_efx_tso_do(txq, added, &m_seg, &in_off, &pend,
					   &pkt_descs, &pkt_len) != 0) {
				/* We may have reached this place for
				 * one of the following reasons:
				 *
				 * 1) Packet header length is greater
				 *    than SFC_TSOH_STD_LEN
				 * 2) TCP header starts at more then
				 *    208 bytes into the frame
				 *
				 * We will deceive RTE saying that we have sent
				 * the packet, but we will actually drop it.
				 * Hence, we should revert 'pend' to the
				 * previous state (in case we have added
				 * VLAN descriptor) and start processing
				 * another one packet. But the original
				 * mbuf shouldn't be orphaned
				 */
				pend -= pkt_descs;

				rte_pktmbuf_free(*pktp);

				continue;
			}

			/*
			 * We've only added 2 FATSOv2 option descriptors
			 * and 1 descriptor for the linearized packet header.
			 * The outstanding work will be done in the same manner
			 * as for the usual non-TSO path
			 */
		}

		for (; m_seg != NULL; m_seg = m_seg->next) {
			efsys_dma_addr_t	next_frag;
			size_t			seg_len;

			seg_len = m_seg->data_len;
			next_frag = rte_mbuf_data_iova(m_seg);

			/*
			 * If we've started TSO transaction few steps earlier,
			 * we'll skip packet header using an offset in the
			 * current segment (which has been set to the
			 * first one containing payload)
			 */
			seg_len -= in_off;
			next_frag += in_off;
			in_off = 0;

			do {
				efsys_dma_addr_t	frag_addr = next_frag;
				size_t			frag_len;

				/*
				 * It is assumed here that there is no
				 * limitation on address boundary
				 * crossing by DMA descriptor.
				 */
				frag_len = MIN(seg_len, txq->dma_desc_size_max);
				next_frag += frag_len;
				seg_len -= frag_len;
				pkt_len -= frag_len;

				efx_tx_qdesc_dma_create(txq->common,
							frag_addr, frag_len,
							(pkt_len == 0),
							pend++);

				pkt_descs++;
			} while (seg_len != 0);
		}

		added += pkt_descs;

		fill_level += pkt_descs;
		if (unlikely(fill_level > hard_max_fill)) {
			/*
			 * Our estimation for maximum number of descriptors
			 * required to send a packet seems to be wrong.
			 * Try to reap (if we haven't yet).
			 */
			if (!reap_done) {
				sfc_efx_tx_reap(txq);
				reap_done = B_TRUE;
				fill_level = added - txq->completed;
				if (fill_level > hard_max_fill) {
					pend -= pkt_descs;
					break;
				}
			} else {
				pend -= pkt_descs;
				break;
			}
		}

		/* Assign mbuf to the last used desc */
		txq->sw_ring[(added - 1) & txq->ptr_mask].mbuf = *pktp;
	}

	if (likely(pkts_sent > 0)) {
		rc = efx_tx_qdesc_post(txq->common, txq->pend_desc,
				       pend - &txq->pend_desc[0],
				       txq->completed, &txq->added);
		SFC_ASSERT(rc == 0);

		if (likely(pushed != txq->added))
			efx_tx_qpush(txq->common, txq->added, pushed);
	}

#if SFC_TX_XMIT_PKTS_REAP_AT_LEAST_ONCE
	if (!reap_done)
		sfc_efx_tx_reap(txq);
#endif

done:
	return pkts_sent;
}

struct sfc_txq *
sfc_txq_by_dp_txq(const struct sfc_dp_txq *dp_txq)
{
	const struct sfc_dp_queue *dpq = &dp_txq->dpq;
	struct rte_eth_dev *eth_dev;
	struct sfc_adapter *sa;
	struct sfc_txq *txq;

	SFC_ASSERT(rte_eth_dev_is_valid_port(dpq->port_id));
	eth_dev = &rte_eth_devices[dpq->port_id];

	sa = eth_dev->data->dev_private;

	SFC_ASSERT(dpq->queue_id < sa->txq_count);
	txq = sa->txq_info[dpq->queue_id].txq;

	SFC_ASSERT(txq != NULL);
	return txq;
}

static sfc_dp_tx_qsize_up_rings_t sfc_efx_tx_qsize_up_rings;
static int
sfc_efx_tx_qsize_up_rings(uint16_t nb_tx_desc,
			  unsigned int *txq_entries,
			  unsigned int *evq_entries,
			  unsigned int *txq_max_fill_level)
{
	*txq_entries = nb_tx_desc;
	*evq_entries = nb_tx_desc;
	*txq_max_fill_level = EFX_TXQ_LIMIT(*txq_entries);
	return 0;
}

static sfc_dp_tx_qcreate_t sfc_efx_tx_qcreate;
static int
sfc_efx_tx_qcreate(uint16_t port_id, uint16_t queue_id,
		   const struct rte_pci_addr *pci_addr,
		   int socket_id,
		   const struct sfc_dp_tx_qcreate_info *info,
		   struct sfc_dp_txq **dp_txqp)
{
	struct sfc_efx_txq *txq;
	struct sfc_txq *ctrl_txq;
	int rc;

	rc = ENOMEM;
	txq = rte_zmalloc_socket("sfc-efx-txq", sizeof(*txq),
				 RTE_CACHE_LINE_SIZE, socket_id);
	if (txq == NULL)
		goto fail_txq_alloc;

	sfc_dp_queue_init(&txq->dp.dpq, port_id, queue_id, pci_addr);

	rc = ENOMEM;
	txq->pend_desc = rte_calloc_socket("sfc-efx-txq-pend-desc",
					   EFX_TXQ_LIMIT(info->txq_entries),
					   sizeof(*txq->pend_desc), 0,
					   socket_id);
	if (txq->pend_desc == NULL)
		goto fail_pend_desc_alloc;

	rc = ENOMEM;
	txq->sw_ring = rte_calloc_socket("sfc-efx-txq-sw_ring",
					 info->txq_entries,
					 sizeof(*txq->sw_ring),
					 RTE_CACHE_LINE_SIZE, socket_id);
	if (txq->sw_ring == NULL)
		goto fail_sw_ring_alloc;

	ctrl_txq = sfc_txq_by_dp_txq(&txq->dp);
	if (ctrl_txq->evq->sa->tso) {
		rc = sfc_efx_tso_alloc_tsoh_objs(txq->sw_ring,
						 info->txq_entries, socket_id);
		if (rc != 0)
			goto fail_alloc_tsoh_objs;
	}

	txq->evq = ctrl_txq->evq;
	txq->ptr_mask = info->txq_entries - 1;
	txq->max_fill_level = info->max_fill_level;
	txq->free_thresh = info->free_thresh;
	txq->dma_desc_size_max = info->dma_desc_size_max;

	*dp_txqp = &txq->dp;
	return 0;

fail_alloc_tsoh_objs:
	rte_free(txq->sw_ring);

fail_sw_ring_alloc:
	rte_free(txq->pend_desc);

fail_pend_desc_alloc:
	rte_free(txq);

fail_txq_alloc:
	return rc;
}

static sfc_dp_tx_qdestroy_t sfc_efx_tx_qdestroy;
static void
sfc_efx_tx_qdestroy(struct sfc_dp_txq *dp_txq)
{
	struct sfc_efx_txq *txq = sfc_efx_txq_by_dp_txq(dp_txq);

	sfc_efx_tso_free_tsoh_objs(txq->sw_ring, txq->ptr_mask + 1);
	rte_free(txq->sw_ring);
	rte_free(txq->pend_desc);
	rte_free(txq);
}

static sfc_dp_tx_qstart_t sfc_efx_tx_qstart;
static int
sfc_efx_tx_qstart(struct sfc_dp_txq *dp_txq,
		  __rte_unused unsigned int evq_read_ptr,
		  unsigned int txq_desc_index)
{
	/* libefx-based datapath is specific to libefx-based PMD */
	struct sfc_efx_txq *txq = sfc_efx_txq_by_dp_txq(dp_txq);
	struct sfc_txq *ctrl_txq = sfc_txq_by_dp_txq(dp_txq);

	txq->common = ctrl_txq->common;

	txq->pending = txq->completed = txq->added = txq_desc_index;
	txq->hw_vlan_tci = 0;

	txq->flags |= (SFC_EFX_TXQ_FLAG_STARTED | SFC_EFX_TXQ_FLAG_RUNNING);

	return 0;
}

static sfc_dp_tx_qstop_t sfc_efx_tx_qstop;
static void
sfc_efx_tx_qstop(struct sfc_dp_txq *dp_txq,
		 __rte_unused unsigned int *evq_read_ptr)
{
	struct sfc_efx_txq *txq = sfc_efx_txq_by_dp_txq(dp_txq);

	txq->flags &= ~SFC_EFX_TXQ_FLAG_RUNNING;
}

static sfc_dp_tx_qreap_t sfc_efx_tx_qreap;
static void
sfc_efx_tx_qreap(struct sfc_dp_txq *dp_txq)
{
	struct sfc_efx_txq *txq = sfc_efx_txq_by_dp_txq(dp_txq);
	unsigned int txds;

	sfc_efx_tx_reap(txq);

	for (txds = 0; txds <= txq->ptr_mask; txds++) {
		if (txq->sw_ring[txds].mbuf != NULL) {
			rte_pktmbuf_free(txq->sw_ring[txds].mbuf);
			txq->sw_ring[txds].mbuf = NULL;
		}
	}

	txq->flags &= ~SFC_EFX_TXQ_FLAG_STARTED;
}

static sfc_dp_tx_qdesc_status_t sfc_efx_tx_qdesc_status;
static int
sfc_efx_tx_qdesc_status(struct sfc_dp_txq *dp_txq, uint16_t offset)
{
	struct sfc_efx_txq *txq = sfc_efx_txq_by_dp_txq(dp_txq);

	if (unlikely(offset > txq->ptr_mask))
		return -EINVAL;

	if (unlikely(offset >= txq->max_fill_level))
		return RTE_ETH_TX_DESC_UNAVAIL;

	/*
	 * Poll EvQ to derive up-to-date 'txq->pending' figure;
	 * it is required for the queue to be running, but the
	 * check is omitted because API design assumes that it
	 * is the duty of the caller to satisfy all conditions
	 */
	SFC_ASSERT((txq->flags & SFC_EFX_TXQ_FLAG_RUNNING) ==
		   SFC_EFX_TXQ_FLAG_RUNNING);
	sfc_ev_qpoll(txq->evq);

	/*
	 * Ring tail is 'txq->pending', and although descriptors
	 * between 'txq->completed' and 'txq->pending' are still
	 * in use by the driver, they should be reported as DONE
	 */
	if (unlikely(offset < (txq->added - txq->pending)))
		return RTE_ETH_TX_DESC_FULL;

	/*
	 * There is no separate return value for unused descriptors;
	 * the latter will be reported as DONE because genuine DONE
	 * descriptors will be freed anyway in SW on the next burst
	 */
	return RTE_ETH_TX_DESC_DONE;
}

struct sfc_dp_tx sfc_efx_tx = {
	.dp = {
		.name		= SFC_KVARG_DATAPATH_EFX,
		.type		= SFC_DP_TX,
		.hw_fw_caps	= 0,
	},
	.features		= SFC_DP_TX_FEAT_VLAN_INSERT |
				  SFC_DP_TX_FEAT_TSO |
				  SFC_DP_TX_FEAT_MULTI_POOL |
				  SFC_DP_TX_FEAT_REFCNT |
				  SFC_DP_TX_FEAT_MULTI_SEG,
	.qsize_up_rings		= sfc_efx_tx_qsize_up_rings,
	.qcreate		= sfc_efx_tx_qcreate,
	.qdestroy		= sfc_efx_tx_qdestroy,
	.qstart			= sfc_efx_tx_qstart,
	.qstop			= sfc_efx_tx_qstop,
	.qreap			= sfc_efx_tx_qreap,
	.qdesc_status		= sfc_efx_tx_qdesc_status,
	.pkt_burst		= sfc_efx_xmit_pkts,
};