summaryrefslogtreecommitdiff
path: root/drivers/net/mlx5/mlx5_rxtx.c
blob: 4c01187faa43ad484e2944749fde46ad26f1211d (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
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
2533
2534
2535
2536
2537
2538
2539
2540
2541
2542
2543
2544
2545
2546
2547
2548
2549
2550
2551
2552
2553
2554
2555
2556
2557
2558
2559
2560
2561
2562
2563
2564
2565
2566
2567
2568
2569
2570
2571
2572
2573
2574
2575
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
2620
2621
2622
2623
2624
2625
2626
2627
2628
2629
2630
2631
2632
2633
2634
2635
2636
2637
2638
2639
2640
2641
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
2669
2670
2671
2672
2673
2674
2675
2676
2677
2678
2679
2680
2681
2682
2683
2684
2685
2686
2687
2688
2689
2690
2691
2692
2693
2694
2695
2696
2697
2698
2699
2700
2701
2702
2703
2704
2705
2706
2707
2708
2709
2710
2711
2712
2713
2714
2715
2716
2717
2718
2719
2720
2721
2722
2723
2724
2725
2726
2727
2728
2729
2730
2731
2732
2733
2734
2735
2736
2737
2738
2739
2740
2741
2742
2743
2744
2745
2746
2747
2748
2749
2750
2751
2752
2753
2754
2755
2756
2757
2758
2759
2760
2761
2762
2763
2764
2765
2766
2767
2768
2769
2770
2771
2772
2773
2774
2775
2776
2777
2778
2779
2780
2781
2782
2783
2784
2785
2786
2787
2788
2789
2790
2791
2792
2793
2794
2795
2796
2797
2798
2799
2800
2801
2802
2803
2804
2805
2806
2807
2808
2809
2810
2811
2812
2813
2814
2815
2816
2817
2818
2819
2820
2821
2822
2823
2824
2825
2826
2827
2828
2829
2830
2831
2832
2833
2834
2835
2836
2837
2838
2839
2840
2841
2842
2843
2844
2845
2846
2847
2848
2849
2850
2851
2852
2853
2854
2855
2856
2857
2858
2859
2860
2861
2862
2863
2864
2865
2866
2867
2868
2869
2870
2871
2872
2873
2874
2875
2876
2877
2878
2879
2880
2881
2882
2883
2884
2885
2886
2887
2888
2889
2890
2891
2892
2893
2894
2895
2896
2897
2898
2899
2900
2901
2902
2903
2904
2905
2906
2907
2908
2909
2910
2911
2912
2913
2914
2915
2916
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926
2927
2928
2929
2930
2931
2932
2933
2934
2935
2936
2937
2938
2939
2940
2941
2942
2943
2944
2945
2946
2947
2948
2949
2950
2951
2952
2953
2954
2955
2956
2957
2958
2959
2960
2961
2962
2963
2964
2965
2966
2967
2968
2969
2970
2971
2972
2973
2974
2975
2976
2977
2978
2979
2980
2981
2982
2983
2984
2985
2986
2987
2988
2989
2990
2991
2992
2993
2994
2995
2996
2997
2998
2999
3000
3001
3002
3003
3004
3005
3006
3007
3008
3009
3010
3011
3012
3013
3014
3015
3016
3017
3018
3019
3020
3021
3022
3023
3024
3025
3026
3027
3028
3029
3030
3031
3032
3033
3034
3035
3036
3037
3038
3039
3040
3041
3042
3043
3044
3045
3046
3047
3048
3049
3050
3051
3052
3053
3054
3055
3056
3057
3058
3059
3060
3061
3062
3063
3064
3065
3066
3067
3068
3069
3070
3071
3072
3073
3074
3075
3076
3077
3078
3079
3080
3081
3082
3083
3084
3085
3086
3087
3088
3089
3090
3091
3092
3093
3094
3095
3096
3097
3098
3099
3100
3101
3102
3103
3104
3105
3106
3107
3108
3109
3110
3111
3112
3113
3114
3115
3116
3117
3118
3119
3120
3121
3122
3123
3124
3125
3126
3127
3128
3129
3130
3131
3132
3133
3134
3135
3136
3137
3138
3139
3140
3141
3142
3143
3144
3145
3146
3147
3148
3149
3150
3151
3152
3153
3154
3155
3156
3157
3158
3159
3160
3161
3162
3163
3164
3165
3166
3167
3168
3169
3170
3171
3172
3173
3174
3175
3176
3177
3178
3179
3180
3181
3182
3183
3184
3185
3186
3187
3188
3189
3190
3191
3192
3193
3194
3195
3196
3197
3198
3199
3200
3201
3202
3203
3204
3205
3206
3207
3208
3209
3210
3211
3212
3213
3214
3215
3216
3217
3218
3219
3220
3221
3222
3223
3224
3225
3226
3227
3228
3229
3230
3231
3232
3233
3234
3235
3236
3237
3238
3239
3240
3241
3242
3243
3244
3245
3246
3247
3248
3249
3250
3251
3252
3253
3254
3255
3256
3257
3258
3259
3260
3261
3262
3263
3264
3265
3266
3267
3268
3269
3270
3271
3272
3273
3274
3275
3276
3277
3278
3279
3280
3281
3282
3283
3284
3285
3286
3287
3288
3289
3290
3291
3292
3293
3294
3295
3296
3297
3298
3299
3300
3301
3302
3303
3304
3305
3306
3307
3308
3309
3310
3311
3312
3313
3314
3315
3316
3317
3318
3319
3320
3321
3322
3323
3324
3325
3326
3327
3328
3329
3330
3331
3332
3333
3334
3335
3336
3337
3338
3339
3340
3341
3342
3343
3344
3345
3346
3347
3348
3349
3350
3351
3352
3353
3354
3355
3356
3357
3358
3359
3360
3361
3362
3363
3364
3365
3366
3367
3368
3369
3370
3371
3372
3373
3374
3375
3376
3377
3378
3379
3380
3381
3382
3383
3384
3385
3386
3387
3388
3389
3390
3391
3392
3393
3394
3395
3396
3397
3398
3399
3400
3401
3402
3403
3404
3405
3406
3407
3408
3409
3410
3411
3412
3413
3414
3415
3416
3417
3418
3419
3420
3421
3422
3423
3424
3425
3426
3427
3428
3429
3430
3431
3432
3433
3434
3435
3436
3437
3438
3439
3440
3441
3442
3443
3444
3445
3446
3447
3448
3449
3450
3451
3452
3453
3454
3455
3456
3457
3458
3459
3460
3461
3462
3463
3464
3465
3466
3467
3468
3469
3470
3471
3472
3473
3474
3475
3476
3477
3478
3479
3480
3481
3482
3483
3484
3485
3486
3487
3488
3489
3490
3491
3492
3493
3494
3495
3496
3497
3498
3499
3500
3501
3502
3503
3504
3505
3506
3507
3508
3509
3510
3511
3512
3513
3514
3515
3516
3517
3518
3519
3520
3521
3522
3523
3524
3525
3526
3527
3528
3529
3530
3531
3532
3533
3534
3535
3536
3537
3538
3539
3540
3541
3542
3543
3544
3545
3546
3547
3548
3549
3550
3551
3552
3553
3554
3555
3556
3557
3558
3559
3560
3561
3562
3563
3564
3565
3566
3567
3568
3569
3570
3571
3572
3573
3574
3575
3576
3577
3578
3579
3580
3581
3582
3583
3584
3585
3586
3587
3588
3589
3590
3591
3592
3593
3594
3595
3596
3597
3598
3599
3600
3601
3602
3603
3604
3605
3606
3607
3608
3609
3610
3611
3612
3613
3614
3615
3616
3617
3618
3619
3620
3621
3622
3623
3624
3625
3626
3627
3628
3629
3630
3631
3632
3633
3634
3635
3636
3637
3638
3639
3640
3641
3642
3643
3644
3645
3646
3647
3648
3649
3650
3651
3652
3653
3654
3655
3656
3657
3658
3659
3660
3661
3662
3663
3664
3665
3666
3667
3668
3669
3670
3671
3672
3673
3674
3675
3676
3677
3678
3679
3680
3681
3682
3683
3684
3685
3686
3687
3688
3689
3690
3691
3692
3693
3694
3695
3696
3697
3698
3699
3700
3701
3702
3703
3704
3705
3706
3707
3708
3709
3710
3711
3712
3713
3714
3715
3716
3717
3718
3719
3720
3721
3722
3723
3724
3725
3726
3727
3728
3729
3730
3731
3732
3733
3734
3735
3736
3737
3738
3739
3740
3741
3742
3743
3744
3745
3746
3747
3748
3749
3750
3751
3752
3753
3754
3755
3756
3757
3758
3759
3760
3761
3762
3763
3764
3765
3766
3767
3768
3769
3770
3771
3772
3773
3774
3775
3776
3777
3778
3779
3780
3781
3782
3783
3784
3785
3786
3787
3788
3789
3790
3791
3792
3793
3794
3795
3796
3797
3798
3799
3800
3801
3802
3803
3804
3805
3806
3807
3808
3809
3810
3811
3812
3813
3814
3815
3816
3817
3818
3819
3820
3821
3822
3823
3824
3825
3826
3827
3828
3829
3830
3831
3832
3833
3834
3835
3836
3837
3838
3839
3840
3841
3842
3843
3844
3845
3846
3847
3848
3849
3850
3851
3852
3853
3854
3855
3856
3857
3858
3859
3860
3861
3862
3863
3864
3865
3866
3867
3868
3869
3870
3871
3872
3873
3874
3875
3876
3877
3878
3879
3880
3881
3882
3883
3884
3885
3886
3887
3888
3889
3890
3891
3892
3893
3894
3895
3896
3897
3898
3899
3900
3901
3902
3903
3904
3905
3906
3907
3908
3909
3910
3911
3912
3913
3914
3915
3916
3917
3918
3919
3920
3921
3922
3923
3924
3925
3926
3927
3928
3929
3930
3931
3932
3933
3934
3935
3936
3937
3938
3939
3940
3941
3942
3943
3944
3945
3946
3947
3948
3949
3950
3951
3952
3953
3954
3955
3956
3957
3958
3959
3960
3961
3962
3963
3964
3965
3966
3967
3968
3969
3970
3971
3972
3973
3974
3975
3976
3977
3978
3979
3980
3981
3982
3983
3984
3985
3986
3987
3988
3989
3990
3991
3992
3993
3994
3995
3996
3997
3998
3999
4000
4001
4002
4003
4004
4005
4006
4007
4008
4009
4010
4011
4012
4013
4014
4015
4016
4017
4018
4019
4020
4021
4022
4023
4024
4025
4026
4027
4028
4029
4030
4031
4032
4033
4034
4035
4036
4037
4038
4039
4040
4041
4042
4043
4044
4045
4046
4047
4048
4049
4050
4051
4052
4053
4054
4055
4056
4057
4058
4059
4060
4061
4062
4063
4064
4065
4066
4067
4068
4069
4070
4071
4072
4073
4074
4075
4076
4077
4078
4079
4080
4081
4082
4083
4084
4085
4086
4087
4088
4089
4090
4091
4092
4093
4094
4095
4096
4097
4098
4099
4100
4101
4102
4103
4104
4105
4106
4107
4108
4109
4110
4111
4112
4113
4114
4115
4116
4117
4118
4119
4120
4121
4122
4123
4124
4125
4126
4127
4128
4129
4130
4131
4132
4133
4134
4135
4136
4137
4138
4139
4140
4141
4142
4143
4144
4145
4146
4147
4148
4149
4150
4151
4152
4153
4154
4155
4156
4157
4158
4159
4160
4161
4162
4163
4164
4165
4166
4167
4168
4169
4170
4171
4172
4173
4174
4175
4176
4177
4178
4179
4180
4181
4182
4183
4184
4185
4186
4187
4188
4189
4190
4191
4192
4193
4194
4195
4196
4197
4198
4199
4200
4201
4202
4203
4204
4205
4206
4207
4208
4209
4210
4211
4212
4213
4214
4215
4216
4217
4218
4219
4220
4221
4222
4223
4224
4225
4226
4227
4228
4229
4230
4231
4232
4233
4234
4235
4236
4237
4238
4239
4240
4241
4242
4243
4244
4245
4246
4247
4248
4249
4250
4251
4252
4253
4254
4255
4256
4257
4258
4259
4260
4261
4262
4263
4264
4265
4266
4267
4268
4269
4270
4271
4272
4273
4274
4275
4276
4277
4278
4279
4280
4281
4282
4283
4284
4285
4286
4287
4288
4289
4290
4291
4292
4293
4294
4295
4296
4297
4298
4299
4300
4301
4302
4303
4304
4305
4306
4307
4308
4309
4310
4311
4312
4313
4314
4315
4316
4317
4318
4319
4320
4321
4322
4323
4324
4325
4326
4327
4328
4329
4330
4331
4332
4333
4334
4335
4336
4337
4338
4339
4340
4341
4342
4343
4344
4345
4346
4347
4348
4349
4350
4351
4352
4353
4354
4355
4356
4357
4358
4359
4360
4361
4362
4363
4364
4365
4366
4367
4368
4369
4370
4371
4372
4373
4374
4375
4376
4377
4378
4379
4380
4381
4382
4383
4384
4385
4386
4387
4388
4389
4390
4391
4392
4393
4394
4395
4396
4397
4398
4399
4400
4401
4402
4403
4404
4405
4406
4407
4408
4409
4410
4411
4412
4413
4414
4415
4416
4417
4418
4419
4420
4421
4422
4423
4424
4425
4426
4427
4428
4429
4430
4431
4432
4433
4434
4435
4436
4437
4438
4439
4440
4441
4442
4443
4444
4445
4446
4447
4448
4449
4450
4451
4452
4453
4454
4455
4456
4457
4458
4459
4460
4461
4462
4463
4464
4465
4466
4467
4468
4469
4470
4471
4472
4473
4474
4475
4476
4477
4478
4479
4480
4481
4482
4483
4484
4485
4486
4487
4488
4489
4490
4491
4492
4493
4494
4495
4496
4497
4498
4499
4500
4501
4502
4503
4504
4505
4506
4507
4508
4509
4510
4511
4512
4513
4514
4515
4516
4517
4518
4519
4520
4521
4522
4523
4524
4525
4526
4527
4528
4529
4530
4531
4532
4533
4534
4535
4536
4537
4538
4539
4540
4541
4542
4543
4544
4545
4546
4547
4548
4549
4550
4551
4552
4553
4554
4555
4556
4557
4558
4559
4560
4561
4562
4563
4564
4565
4566
4567
4568
4569
4570
4571
4572
4573
4574
4575
4576
4577
4578
4579
4580
4581
4582
4583
4584
4585
4586
4587
4588
4589
4590
4591
4592
4593
4594
4595
4596
4597
4598
4599
4600
4601
4602
4603
4604
4605
4606
4607
4608
4609
4610
4611
4612
4613
4614
4615
4616
4617
4618
4619
4620
4621
4622
4623
4624
4625
4626
4627
4628
4629
4630
4631
4632
4633
4634
4635
4636
4637
4638
4639
4640
4641
4642
4643
4644
4645
4646
4647
4648
4649
4650
4651
4652
4653
4654
4655
4656
4657
4658
4659
4660
4661
4662
4663
4664
4665
4666
4667
4668
4669
4670
4671
4672
4673
4674
4675
4676
4677
4678
4679
4680
4681
4682
4683
4684
4685
4686
4687
4688
4689
4690
4691
4692
4693
4694
4695
4696
4697
4698
4699
4700
4701
4702
4703
4704
4705
4706
4707
4708
4709
4710
4711
4712
4713
4714
4715
4716
4717
4718
4719
4720
4721
4722
4723
4724
4725
4726
4727
4728
4729
4730
4731
4732
4733
4734
4735
4736
4737
4738
4739
4740
4741
4742
4743
4744
4745
4746
4747
4748
4749
4750
4751
4752
4753
4754
4755
4756
4757
4758
4759
4760
4761
4762
4763
4764
4765
4766
4767
4768
4769
4770
4771
4772
4773
4774
4775
4776
4777
4778
4779
4780
4781
4782
4783
4784
4785
4786
4787
4788
4789
4790
4791
4792
4793
4794
4795
4796
4797
4798
4799
4800
4801
4802
4803
4804
4805
4806
4807
4808
4809
4810
4811
4812
4813
4814
4815
4816
4817
4818
4819
4820
4821
4822
4823
4824
4825
4826
4827
4828
4829
4830
4831
4832
4833
4834
4835
4836
4837
4838
4839
4840
4841
4842
4843
4844
4845
4846
4847
4848
4849
4850
4851
4852
4853
4854
4855
4856
4857
4858
4859
4860
4861
4862
4863
4864
4865
4866
4867
4868
4869
4870
4871
4872
4873
4874
4875
4876
4877
4878
4879
4880
4881
4882
4883
4884
4885
4886
4887
4888
4889
4890
4891
4892
4893
4894
4895
4896
4897
4898
4899
4900
4901
4902
4903
4904
4905
4906
4907
4908
4909
4910
4911
4912
4913
4914
4915
4916
4917
4918
4919
4920
4921
4922
4923
4924
4925
4926
4927
4928
4929
4930
4931
4932
4933
4934
4935
4936
4937
4938
4939
4940
4941
4942
4943
4944
4945
4946
4947
4948
4949
4950
4951
4952
4953
4954
4955
4956
4957
4958
4959
4960
4961
4962
4963
4964
4965
4966
4967
4968
4969
4970
4971
4972
4973
4974
4975
4976
4977
4978
4979
4980
4981
4982
4983
4984
4985
4986
4987
4988
4989
4990
4991
4992
4993
4994
4995
4996
4997
4998
4999
5000
5001
5002
5003
5004
5005
5006
5007
5008
5009
5010
5011
5012
5013
5014
5015
5016
5017
5018
5019
5020
5021
5022
5023
5024
5025
5026
5027
5028
5029
5030
5031
5032
5033
5034
5035
5036
5037
5038
5039
5040
5041
5042
5043
5044
5045
5046
5047
5048
5049
5050
5051
5052
5053
5054
5055
5056
5057
5058
5059
5060
5061
5062
5063
5064
5065
5066
5067
5068
5069
5070
5071
5072
5073
5074
5075
5076
5077
5078
5079
5080
5081
5082
5083
5084
5085
5086
5087
5088
5089
5090
5091
5092
5093
5094
5095
5096
5097
5098
5099
5100
5101
5102
5103
5104
5105
5106
5107
5108
5109
5110
5111
5112
5113
5114
5115
5116
5117
5118
5119
5120
5121
5122
5123
5124
5125
5126
5127
5128
5129
5130
5131
5132
5133
5134
5135
5136
5137
5138
5139
5140
5141
5142
5143
5144
5145
5146
5147
5148
5149
5150
5151
5152
5153
5154
5155
5156
5157
5158
5159
5160
5161
5162
5163
5164
5165
5166
5167
5168
5169
5170
5171
5172
5173
5174
5175
5176
5177
5178
5179
5180
5181
5182
5183
5184
5185
5186
5187
5188
5189
5190
5191
5192
5193
5194
5195
5196
5197
5198
5199
5200
5201
5202
5203
5204
5205
5206
5207
5208
/* SPDX-License-Identifier: BSD-3-Clause
 * Copyright 2015 6WIND S.A.
 * Copyright 2015-2019 Mellanox Technologies, Ltd
 */

#include <assert.h>
#include <stdint.h>
#include <string.h>
#include <stdlib.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/mlx5dv.h>
#ifdef PEDANTIC
#pragma GCC diagnostic error "-Wpedantic"
#endif

#include <rte_mbuf.h>
#include <rte_mempool.h>
#include <rte_prefetch.h>
#include <rte_common.h>
#include <rte_branch_prediction.h>
#include <rte_ether.h>
#include <rte_cycles.h>

#include "mlx5.h"
#include "mlx5_utils.h"
#include "mlx5_rxtx.h"
#include "mlx5_autoconf.h"
#include "mlx5_defs.h"
#include "mlx5_prm.h"

/* TX burst subroutines return codes. */
enum mlx5_txcmp_code {
	MLX5_TXCMP_CODE_EXIT = 0,
	MLX5_TXCMP_CODE_ERROR,
	MLX5_TXCMP_CODE_SINGLE,
	MLX5_TXCMP_CODE_MULTI,
	MLX5_TXCMP_CODE_TSO,
	MLX5_TXCMP_CODE_EMPW,
};

/*
 * These defines are used to configure Tx burst routine option set
 * supported at compile time. The not specified options are optimized out
 * out due to if conditions can be explicitly calculated at compile time.
 * The offloads with bigger runtime check (require more CPU cycles to
 * skip) overhead should have the bigger index - this is needed to
 * select the better matching routine function if no exact match and
 * some offloads are not actually requested.
 */
#define MLX5_TXOFF_CONFIG_MULTI (1u << 0) /* Multi-segment packets.*/
#define MLX5_TXOFF_CONFIG_TSO (1u << 1) /* TCP send offload supported.*/
#define MLX5_TXOFF_CONFIG_SWP (1u << 2) /* Tunnels/SW Parser offloads.*/
#define MLX5_TXOFF_CONFIG_CSUM (1u << 3) /* Check Sums offloaded. */
#define MLX5_TXOFF_CONFIG_INLINE (1u << 4) /* Data inlining supported. */
#define MLX5_TXOFF_CONFIG_VLAN (1u << 5) /* VLAN insertion supported.*/
#define MLX5_TXOFF_CONFIG_METADATA (1u << 6) /* Flow metadata. */
#define MLX5_TXOFF_CONFIG_EMPW (1u << 8) /* Enhanced MPW supported.*/

/* The most common offloads groups. */
#define MLX5_TXOFF_CONFIG_NONE 0
#define MLX5_TXOFF_CONFIG_FULL (MLX5_TXOFF_CONFIG_MULTI | \
				MLX5_TXOFF_CONFIG_TSO | \
				MLX5_TXOFF_CONFIG_SWP | \
				MLX5_TXOFF_CONFIG_CSUM | \
				MLX5_TXOFF_CONFIG_INLINE | \
				MLX5_TXOFF_CONFIG_VLAN | \
				MLX5_TXOFF_CONFIG_METADATA)

#define MLX5_TXOFF_CONFIG(mask) (olx & MLX5_TXOFF_CONFIG_##mask)

#define MLX5_TXOFF_DECL(func, olx) \
static uint16_t mlx5_tx_burst_##func(void *txq, \
				     struct rte_mbuf **pkts, \
				    uint16_t pkts_n) \
{ \
	return mlx5_tx_burst_tmpl((struct mlx5_txq_data *)txq, \
		    pkts, pkts_n, (olx)); \
}

#define MLX5_TXOFF_INFO(func, olx) {mlx5_tx_burst_##func, olx},

static __rte_always_inline uint32_t
rxq_cq_to_pkt_type(struct mlx5_rxq_data *rxq, volatile struct mlx5_cqe *cqe);

static __rte_always_inline int
mlx5_rx_poll_len(struct mlx5_rxq_data *rxq, volatile struct mlx5_cqe *cqe,
		 uint16_t cqe_cnt, volatile struct mlx5_mini_cqe8 **mcqe);

static __rte_always_inline uint32_t
rxq_cq_to_ol_flags(volatile struct mlx5_cqe *cqe);

static __rte_always_inline void
rxq_cq_to_mbuf(struct mlx5_rxq_data *rxq, struct rte_mbuf *pkt,
	       volatile struct mlx5_cqe *cqe, uint32_t rss_hash_res);

static __rte_always_inline void
mprq_buf_replace(struct mlx5_rxq_data *rxq, uint16_t rq_idx,
		 const unsigned int strd_n);

static int
mlx5_queue_state_modify(struct rte_eth_dev *dev,
			struct mlx5_mp_arg_queue_state_modify *sm);

static inline void
mlx5_lro_update_tcp_hdr(struct rte_tcp_hdr *restrict tcp,
			volatile struct mlx5_cqe *restrict cqe,
			uint32_t phcsum);

static inline void
mlx5_lro_update_hdr(uint8_t *restrict padd,
		    volatile struct mlx5_cqe *restrict cqe,
		    uint32_t len);

uint32_t mlx5_ptype_table[] __rte_cache_aligned = {
	[0xff] = RTE_PTYPE_ALL_MASK, /* Last entry for errored packet. */
};

uint8_t mlx5_cksum_table[1 << 10] __rte_cache_aligned;
uint8_t mlx5_swp_types_table[1 << 10] __rte_cache_aligned;

/**
 * Build a table to translate Rx completion flags to packet type.
 *
 * @note: fix mlx5_dev_supported_ptypes_get() if any change here.
 */
void
mlx5_set_ptype_table(void)
{
	unsigned int i;
	uint32_t (*p)[RTE_DIM(mlx5_ptype_table)] = &mlx5_ptype_table;

	/* Last entry must not be overwritten, reserved for errored packet. */
	for (i = 0; i < RTE_DIM(mlx5_ptype_table) - 1; ++i)
		(*p)[i] = RTE_PTYPE_UNKNOWN;
	/*
	 * The index to the array should have:
	 * bit[1:0] = l3_hdr_type
	 * bit[4:2] = l4_hdr_type
	 * bit[5] = ip_frag
	 * bit[6] = tunneled
	 * bit[7] = outer_l3_type
	 */
	/* L2 */
	(*p)[0x00] = RTE_PTYPE_L2_ETHER;
	/* L3 */
	(*p)[0x01] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_L4_NONFRAG;
	(*p)[0x02] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_L4_NONFRAG;
	/* Fragmented */
	(*p)[0x21] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_L4_FRAG;
	(*p)[0x22] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_L4_FRAG;
	/* TCP */
	(*p)[0x05] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_L4_TCP;
	(*p)[0x06] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_L4_TCP;
	(*p)[0x0d] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_L4_TCP;
	(*p)[0x0e] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_L4_TCP;
	(*p)[0x11] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_L4_TCP;
	(*p)[0x12] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_L4_TCP;
	/* UDP */
	(*p)[0x09] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_L4_UDP;
	(*p)[0x0a] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_L4_UDP;
	/* Repeat with outer_l3_type being set. Just in case. */
	(*p)[0x81] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_L4_NONFRAG;
	(*p)[0x82] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_L4_NONFRAG;
	(*p)[0xa1] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_L4_FRAG;
	(*p)[0xa2] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_L4_FRAG;
	(*p)[0x85] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_L4_TCP;
	(*p)[0x86] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_L4_TCP;
	(*p)[0x8d] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_L4_TCP;
	(*p)[0x8e] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_L4_TCP;
	(*p)[0x91] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_L4_TCP;
	(*p)[0x92] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_L4_TCP;
	(*p)[0x89] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_L4_UDP;
	(*p)[0x8a] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_L4_UDP;
	/* Tunneled - L3 */
	(*p)[0x40] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
	(*p)[0x41] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L4_NONFRAG;
	(*p)[0x42] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L4_NONFRAG;
	(*p)[0xc0] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
	(*p)[0xc1] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L4_NONFRAG;
	(*p)[0xc2] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L4_NONFRAG;
	/* Tunneled - Fragmented */
	(*p)[0x61] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L4_FRAG;
	(*p)[0x62] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L4_FRAG;
	(*p)[0xe1] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L4_FRAG;
	(*p)[0xe2] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L4_FRAG;
	/* Tunneled - TCP */
	(*p)[0x45] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L4_TCP;
	(*p)[0x46] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L4_TCP;
	(*p)[0x4d] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L4_TCP;
	(*p)[0x4e] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L4_TCP;
	(*p)[0x51] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L4_TCP;
	(*p)[0x52] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L4_TCP;
	(*p)[0xc5] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L4_TCP;
	(*p)[0xc6] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L4_TCP;
	(*p)[0xcd] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L4_TCP;
	(*p)[0xce] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L4_TCP;
	(*p)[0xd1] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L4_TCP;
	(*p)[0xd2] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L4_TCP;
	/* Tunneled - UDP */
	(*p)[0x49] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L4_UDP;
	(*p)[0x4a] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L4_UDP;
	(*p)[0xc9] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L4_UDP;
	(*p)[0xca] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN |
		     RTE_PTYPE_INNER_L4_UDP;
}

/**
 * Build a table to translate packet to checksum type of Verbs.
 */
void
mlx5_set_cksum_table(void)
{
	unsigned int i;
	uint8_t v;

	/*
	 * The index should have:
	 * bit[0] = PKT_TX_TCP_SEG
	 * bit[2:3] = PKT_TX_UDP_CKSUM, PKT_TX_TCP_CKSUM
	 * bit[4] = PKT_TX_IP_CKSUM
	 * bit[8] = PKT_TX_OUTER_IP_CKSUM
	 * bit[9] = tunnel
	 */
	for (i = 0; i < RTE_DIM(mlx5_cksum_table); ++i) {
		v = 0;
		if (i & (1 << 9)) {
			/* Tunneled packet. */
			if (i & (1 << 8)) /* Outer IP. */
				v |= MLX5_ETH_WQE_L3_CSUM;
			if (i & (1 << 4)) /* Inner IP. */
				v |= MLX5_ETH_WQE_L3_INNER_CSUM;
			if (i & (3 << 2 | 1 << 0)) /* L4 or TSO. */
				v |= MLX5_ETH_WQE_L4_INNER_CSUM;
		} else {
			/* No tunnel. */
			if (i & (1 << 4)) /* IP. */
				v |= MLX5_ETH_WQE_L3_CSUM;
			if (i & (3 << 2 | 1 << 0)) /* L4 or TSO. */
				v |= MLX5_ETH_WQE_L4_CSUM;
		}
		mlx5_cksum_table[i] = v;
	}
}

/**
 * Build a table to translate packet type of mbuf to SWP type of Verbs.
 */
void
mlx5_set_swp_types_table(void)
{
	unsigned int i;
	uint8_t v;

	/*
	 * The index should have:
	 * bit[0:1] = PKT_TX_L4_MASK
	 * bit[4] = PKT_TX_IPV6
	 * bit[8] = PKT_TX_OUTER_IPV6
	 * bit[9] = PKT_TX_OUTER_UDP
	 */
	for (i = 0; i < RTE_DIM(mlx5_swp_types_table); ++i) {
		v = 0;
		if (i & (1 << 8))
			v |= MLX5_ETH_WQE_L3_OUTER_IPV6;
		if (i & (1 << 9))
			v |= MLX5_ETH_WQE_L4_OUTER_UDP;
		if (i & (1 << 4))
			v |= MLX5_ETH_WQE_L3_INNER_IPV6;
		if ((i & 3) == (PKT_TX_UDP_CKSUM >> 52))
			v |= MLX5_ETH_WQE_L4_INNER_UDP;
		mlx5_swp_types_table[i] = v;
	}
}

/**
 * Set Software Parser flags and offsets in Ethernet Segment of WQE.
 * Flags must be preliminary initialized to zero.
 *
 * @param loc
 *   Pointer to burst routine local context.
 * @param swp_flags
 *   Pointer to store Software Parser flags
 * @param olx
 *   Configured Tx offloads mask. It is fully defined at
 *   compile time and may be used for optimization.
 *
 * @return
 *   Software Parser offsets packed in dword.
 *   Software Parser flags are set by pointer.
 */
static __rte_always_inline uint32_t
txq_mbuf_to_swp(struct mlx5_txq_local *restrict loc,
		uint8_t *swp_flags,
		unsigned int olx)
{
	uint64_t ol, tunnel;
	unsigned int idx, off;
	uint32_t set;

	if (!MLX5_TXOFF_CONFIG(SWP))
		return 0;
	ol = loc->mbuf->ol_flags;
	tunnel = ol & PKT_TX_TUNNEL_MASK;
	/*
	 * Check whether Software Parser is required.
	 * Only customized tunnels may ask for.
	 */
	if (likely(tunnel != PKT_TX_TUNNEL_UDP && tunnel != PKT_TX_TUNNEL_IP))
		return 0;
	/*
	 * The index should have:
	 * bit[0:1] = PKT_TX_L4_MASK
	 * bit[4] = PKT_TX_IPV6
	 * bit[8] = PKT_TX_OUTER_IPV6
	 * bit[9] = PKT_TX_OUTER_UDP
	 */
	idx = (ol & (PKT_TX_L4_MASK | PKT_TX_IPV6 | PKT_TX_OUTER_IPV6)) >> 52;
	idx |= (tunnel == PKT_TX_TUNNEL_UDP) ? (1 << 9) : 0;
	*swp_flags = mlx5_swp_types_table[idx];
	/*
	 * Set offsets for SW parser. Since ConnectX-5, SW parser just
	 * complements HW parser. SW parser starts to engage only if HW parser
	 * can't reach a header. For the older devices, HW parser will not kick
	 * in if any of SWP offsets is set. Therefore, all of the L3 offsets
	 * should be set regardless of HW offload.
	 */
	off = loc->mbuf->outer_l2_len;
	if (MLX5_TXOFF_CONFIG(VLAN) && ol & PKT_TX_VLAN_PKT)
		off += sizeof(struct rte_vlan_hdr);
	set = (off >> 1) << 8; /* Outer L3 offset. */
	off += loc->mbuf->outer_l3_len;
	if (tunnel == PKT_TX_TUNNEL_UDP)
		set |= off >> 1; /* Outer L4 offset. */
	if (ol & (PKT_TX_IPV4 | PKT_TX_IPV6)) { /* Inner IP. */
		const uint64_t csum = ol & PKT_TX_L4_MASK;
			off += loc->mbuf->l2_len;
		set |= (off >> 1) << 24; /* Inner L3 offset. */
		if (csum == PKT_TX_TCP_CKSUM ||
		    csum == PKT_TX_UDP_CKSUM ||
		    (MLX5_TXOFF_CONFIG(TSO) && ol & PKT_TX_TCP_SEG)) {
			off += loc->mbuf->l3_len;
			set |= (off >> 1) << 16; /* Inner L4 offset. */
		}
	}
	set = rte_cpu_to_le_32(set);
	return set;
}

/**
 * Convert the Checksum offloads to Verbs.
 *
 * @param buf
 *   Pointer to the mbuf.
 *
 * @return
 *   Converted checksum flags.
 */
static __rte_always_inline uint8_t
txq_ol_cksum_to_cs(struct rte_mbuf *buf)
{
	uint32_t idx;
	uint8_t is_tunnel = !!(buf->ol_flags & PKT_TX_TUNNEL_MASK);
	const uint64_t ol_flags_mask = PKT_TX_TCP_SEG | PKT_TX_L4_MASK |
				       PKT_TX_IP_CKSUM | PKT_TX_OUTER_IP_CKSUM;

	/*
	 * The index should have:
	 * bit[0] = PKT_TX_TCP_SEG
	 * bit[2:3] = PKT_TX_UDP_CKSUM, PKT_TX_TCP_CKSUM
	 * bit[4] = PKT_TX_IP_CKSUM
	 * bit[8] = PKT_TX_OUTER_IP_CKSUM
	 * bit[9] = tunnel
	 */
	idx = ((buf->ol_flags & ol_flags_mask) >> 50) | (!!is_tunnel << 9);
	return mlx5_cksum_table[idx];
}

/**
 * Internal function to compute the number of used descriptors in an RX queue
 *
 * @param rxq
 *   The Rx queue.
 *
 * @return
 *   The number of used rx descriptor.
 */
static uint32_t
rx_queue_count(struct mlx5_rxq_data *rxq)
{
	struct rxq_zip *zip = &rxq->zip;
	volatile struct mlx5_cqe *cqe;
	const unsigned int cqe_n = (1 << rxq->cqe_n);
	const unsigned int cqe_cnt = cqe_n - 1;
	unsigned int cq_ci;
	unsigned int used;

	/* if we are processing a compressed cqe */
	if (zip->ai) {
		used = zip->cqe_cnt - zip->ca;
		cq_ci = zip->cq_ci;
	} else {
		used = 0;
		cq_ci = rxq->cq_ci;
	}
	cqe = &(*rxq->cqes)[cq_ci & cqe_cnt];
	while (check_cqe(cqe, cqe_n, cq_ci) != MLX5_CQE_STATUS_HW_OWN) {
		int8_t op_own;
		unsigned int n;

		op_own = cqe->op_own;
		if (MLX5_CQE_FORMAT(op_own) == MLX5_COMPRESSED)
			n = rte_be_to_cpu_32(cqe->byte_cnt);
		else
			n = 1;
		cq_ci += n;
		used += n;
		cqe = &(*rxq->cqes)[cq_ci & cqe_cnt];
	}
	used = RTE_MIN(used, (1U << rxq->elts_n) - 1);
	return used;
}

/**
 * DPDK callback to check the status of a rx descriptor.
 *
 * @param rx_queue
 *   The Rx queue.
 * @param[in] offset
 *   The index of the descriptor in the ring.
 *
 * @return
 *   The status of the tx descriptor.
 */
int
mlx5_rx_descriptor_status(void *rx_queue, uint16_t offset)
{
	struct mlx5_rxq_data *rxq = rx_queue;
	struct mlx5_rxq_ctrl *rxq_ctrl =
			container_of(rxq, struct mlx5_rxq_ctrl, rxq);
	struct rte_eth_dev *dev = ETH_DEV(rxq_ctrl->priv);

	if (dev->rx_pkt_burst != mlx5_rx_burst) {
		rte_errno = ENOTSUP;
		return -rte_errno;
	}
	if (offset >= (1 << rxq->elts_n)) {
		rte_errno = EINVAL;
		return -rte_errno;
	}
	if (offset < rx_queue_count(rxq))
		return RTE_ETH_RX_DESC_DONE;
	return RTE_ETH_RX_DESC_AVAIL;
}

/**
 * DPDK callback to get the number of used descriptors in a RX queue
 *
 * @param dev
 *   Pointer to the device structure.
 *
 * @param rx_queue_id
 *   The Rx queue.
 *
 * @return
 *   The number of used rx descriptor.
 *   -EINVAL if the queue is invalid
 */
uint32_t
mlx5_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
{
	struct mlx5_priv *priv = dev->data->dev_private;
	struct mlx5_rxq_data *rxq;

	if (dev->rx_pkt_burst != mlx5_rx_burst) {
		rte_errno = ENOTSUP;
		return -rte_errno;
	}
	rxq = (*priv->rxqs)[rx_queue_id];
	if (!rxq) {
		rte_errno = EINVAL;
		return -rte_errno;
	}
	return rx_queue_count(rxq);
}

#define MLX5_SYSTEM_LOG_DIR "/var/log"
/**
 * Dump debug information to log file.
 *
 * @param fname
 *   The file name.
 * @param hex_title
 *   If not NULL this string is printed as a header to the output
 *   and the output will be in hexadecimal view.
 * @param buf
 *   This is the buffer address to print out.
 * @param len
 *   The number of bytes to dump out.
 */
void
mlx5_dump_debug_information(const char *fname, const char *hex_title,
			    const void *buf, unsigned int hex_len)
{
	FILE *fd;

	MKSTR(path, "%s/%s", MLX5_SYSTEM_LOG_DIR, fname);
	fd = fopen(path, "a+");
	if (!fd) {
		DRV_LOG(WARNING, "cannot open %s for debug dump\n",
			path);
		MKSTR(path2, "./%s", fname);
		fd = fopen(path2, "a+");
		if (!fd) {
			DRV_LOG(ERR, "cannot open %s for debug dump\n",
				path2);
			return;
		}
		DRV_LOG(INFO, "New debug dump in file %s\n", path2);
	} else {
		DRV_LOG(INFO, "New debug dump in file %s\n", path);
	}
	if (hex_title)
		rte_hexdump(fd, hex_title, buf, hex_len);
	else
		fprintf(fd, "%s", (const char *)buf);
	fprintf(fd, "\n\n\n");
	fclose(fd);
}

/**
 * Move QP from error state to running state and initialize indexes.
 *
 * @param txq_ctrl
 *   Pointer to TX queue control structure.
 *
 * @return
 *   0 on success, else -1.
 */
static int
tx_recover_qp(struct mlx5_txq_ctrl *txq_ctrl)
{
	struct mlx5_mp_arg_queue_state_modify sm = {
			.is_wq = 0,
			.queue_id = txq_ctrl->txq.idx,
	};

	if (mlx5_queue_state_modify(ETH_DEV(txq_ctrl->priv), &sm))
		return -1;
	txq_ctrl->txq.wqe_ci = 0;
	txq_ctrl->txq.wqe_pi = 0;
	txq_ctrl->txq.elts_comp = 0;
	return 0;
}

/* Return 1 if the error CQE is signed otherwise, sign it and return 0. */
static int
check_err_cqe_seen(volatile struct mlx5_err_cqe *err_cqe)
{
	static const uint8_t magic[] = "seen";
	int ret = 1;
	unsigned int i;

	for (i = 0; i < sizeof(magic); ++i)
		if (!ret || err_cqe->rsvd1[i] != magic[i]) {
			ret = 0;
			err_cqe->rsvd1[i] = magic[i];
		}
	return ret;
}

/**
 * Handle error CQE.
 *
 * @param txq
 *   Pointer to TX queue structure.
 * @param error_cqe
 *   Pointer to the error CQE.
 *
 * @return
 *   Negative value if queue recovery failed,
 *   the last Tx buffer element to free otherwise.
 */
int
mlx5_tx_error_cqe_handle(struct mlx5_txq_data *restrict txq,
			 volatile struct mlx5_err_cqe *err_cqe)
{
	if (err_cqe->syndrome != MLX5_CQE_SYNDROME_WR_FLUSH_ERR) {
		const uint16_t wqe_m = ((1 << txq->wqe_n) - 1);
		struct mlx5_txq_ctrl *txq_ctrl =
				container_of(txq, struct mlx5_txq_ctrl, txq);
		uint16_t new_wqe_pi = rte_be_to_cpu_16(err_cqe->wqe_counter);
		int seen = check_err_cqe_seen(err_cqe);

		if (!seen && txq_ctrl->dump_file_n <
		    txq_ctrl->priv->config.max_dump_files_num) {
			MKSTR(err_str, "Unexpected CQE error syndrome "
			      "0x%02x CQN = %u SQN = %u wqe_counter = %u "
			      "wq_ci = %u cq_ci = %u", err_cqe->syndrome,
			      txq->cqe_s, txq->qp_num_8s >> 8,
			      rte_be_to_cpu_16(err_cqe->wqe_counter),
			      txq->wqe_ci, txq->cq_ci);
			MKSTR(name, "dpdk_mlx5_port_%u_txq_%u_index_%u_%u",
			      PORT_ID(txq_ctrl->priv), txq->idx,
			      txq_ctrl->dump_file_n, (uint32_t)rte_rdtsc());
			mlx5_dump_debug_information(name, NULL, err_str, 0);
			mlx5_dump_debug_information(name, "MLX5 Error CQ:",
						    (const void *)((uintptr_t)
						    txq->cqes),
						    sizeof(*err_cqe) *
						    (1 << txq->cqe_n));
			mlx5_dump_debug_information(name, "MLX5 Error SQ:",
						    (const void *)((uintptr_t)
						    txq->wqes),
						    MLX5_WQE_SIZE *
						    (1 << txq->wqe_n));
			txq_ctrl->dump_file_n++;
		}
		if (!seen)
			/*
			 * Count errors in WQEs units.
			 * Later it can be improved to count error packets,
			 * for example, by SQ parsing to find how much packets
			 * should be counted for each WQE.
			 */
			txq->stats.oerrors += ((txq->wqe_ci & wqe_m) -
						new_wqe_pi) & wqe_m;
		if (tx_recover_qp(txq_ctrl) == 0) {
			txq->cq_ci++;
			/* Release all the remaining buffers. */
			return txq->elts_head;
		}
		/* Recovering failed - try again later on the same WQE. */
		return -1;
	} else {
		txq->cq_ci++;
	}
	/* Do not release buffers. */
	return txq->elts_tail;
}

/**
 * Translate RX completion flags to packet type.
 *
 * @param[in] rxq
 *   Pointer to RX queue structure.
 * @param[in] cqe
 *   Pointer to CQE.
 *
 * @note: fix mlx5_dev_supported_ptypes_get() if any change here.
 *
 * @return
 *   Packet type for struct rte_mbuf.
 */
static inline uint32_t
rxq_cq_to_pkt_type(struct mlx5_rxq_data *rxq, volatile struct mlx5_cqe *cqe)
{
	uint8_t idx;
	uint8_t pinfo = cqe->pkt_info;
	uint16_t ptype = cqe->hdr_type_etc;

	/*
	 * The index to the array should have:
	 * bit[1:0] = l3_hdr_type
	 * bit[4:2] = l4_hdr_type
	 * bit[5] = ip_frag
	 * bit[6] = tunneled
	 * bit[7] = outer_l3_type
	 */
	idx = ((pinfo & 0x3) << 6) | ((ptype & 0xfc00) >> 10);
	return mlx5_ptype_table[idx] | rxq->tunnel * !!(idx & (1 << 6));
}

/**
 * Initialize Rx WQ and indexes.
 *
 * @param[in] rxq
 *   Pointer to RX queue structure.
 */
void
mlx5_rxq_initialize(struct mlx5_rxq_data *rxq)
{
	const unsigned int wqe_n = 1 << rxq->elts_n;
	unsigned int i;

	for (i = 0; (i != wqe_n); ++i) {
		volatile struct mlx5_wqe_data_seg *scat;
		uintptr_t addr;
		uint32_t byte_count;

		if (mlx5_rxq_mprq_enabled(rxq)) {
			struct mlx5_mprq_buf *buf = (*rxq->mprq_bufs)[i];

			scat = &((volatile struct mlx5_wqe_mprq *)
				rxq->wqes)[i].dseg;
			addr = (uintptr_t)mlx5_mprq_buf_addr(buf,
							 1 << rxq->strd_num_n);
			byte_count = (1 << rxq->strd_sz_n) *
					(1 << rxq->strd_num_n);
		} else {
			struct rte_mbuf *buf = (*rxq->elts)[i];

			scat = &((volatile struct mlx5_wqe_data_seg *)
					rxq->wqes)[i];
			addr = rte_pktmbuf_mtod(buf, uintptr_t);
			byte_count = DATA_LEN(buf);
		}
		/* scat->addr must be able to store a pointer. */
		assert(sizeof(scat->addr) >= sizeof(uintptr_t));
		*scat = (struct mlx5_wqe_data_seg){
			.addr = rte_cpu_to_be_64(addr),
			.byte_count = rte_cpu_to_be_32(byte_count),
			.lkey = mlx5_rx_addr2mr(rxq, addr),
		};
	}
	rxq->consumed_strd = 0;
	rxq->decompressed = 0;
	rxq->rq_pi = 0;
	rxq->zip = (struct rxq_zip){
		.ai = 0,
	};
	/* Update doorbell counter. */
	rxq->rq_ci = wqe_n >> rxq->sges_n;
	rte_cio_wmb();
	*rxq->rq_db = rte_cpu_to_be_32(rxq->rq_ci);
}

/**
 * Modify a Verbs/DevX queue state.
 * This must be called from the primary process.
 *
 * @param dev
 *   Pointer to Ethernet device.
 * @param sm
 *   State modify request parameters.
 *
 * @return
 *   0 in case of success else non-zero value and rte_errno is set.
 */
int
mlx5_queue_state_modify_primary(struct rte_eth_dev *dev,
			const struct mlx5_mp_arg_queue_state_modify *sm)
{
	int ret;
	struct mlx5_priv *priv = dev->data->dev_private;

	if (sm->is_wq) {
		struct mlx5_rxq_data *rxq = (*priv->rxqs)[sm->queue_id];
		struct mlx5_rxq_ctrl *rxq_ctrl =
			container_of(rxq, struct mlx5_rxq_ctrl, rxq);

		if (rxq_ctrl->obj->type == MLX5_RXQ_OBJ_TYPE_IBV) {
			struct ibv_wq_attr mod = {
				.attr_mask = IBV_WQ_ATTR_STATE,
				.wq_state = sm->state,
			};

			ret = mlx5_glue->modify_wq(rxq_ctrl->obj->wq, &mod);
		} else { /* rxq_ctrl->obj->type == MLX5_RXQ_OBJ_TYPE_DEVX_RQ. */
			struct mlx5_devx_modify_rq_attr rq_attr;

			memset(&rq_attr, 0, sizeof(rq_attr));
			if (sm->state == IBV_WQS_RESET) {
				rq_attr.rq_state = MLX5_RQC_STATE_ERR;
				rq_attr.state = MLX5_RQC_STATE_RST;
			} else if (sm->state == IBV_WQS_RDY) {
				rq_attr.rq_state = MLX5_RQC_STATE_RST;
				rq_attr.state = MLX5_RQC_STATE_RDY;
			} else if (sm->state == IBV_WQS_ERR) {
				rq_attr.rq_state = MLX5_RQC_STATE_RDY;
				rq_attr.state = MLX5_RQC_STATE_ERR;
			}
			ret = mlx5_devx_cmd_modify_rq(rxq_ctrl->obj->rq,
						      &rq_attr);
		}
		if (ret) {
			DRV_LOG(ERR, "Cannot change Rx WQ state to %u  - %s\n",
					sm->state, strerror(errno));
			rte_errno = errno;
			return ret;
		}
	} else {
		struct mlx5_txq_data *txq = (*priv->txqs)[sm->queue_id];
		struct mlx5_txq_ctrl *txq_ctrl =
			container_of(txq, struct mlx5_txq_ctrl, txq);
		struct ibv_qp_attr mod = {
			.qp_state = IBV_QPS_RESET,
			.port_num = (uint8_t)priv->ibv_port,
		};
		struct ibv_qp *qp = txq_ctrl->ibv->qp;

		ret = mlx5_glue->modify_qp(qp, &mod, IBV_QP_STATE);
		if (ret) {
			DRV_LOG(ERR, "Cannot change the Tx QP state to RESET "
				"%s\n", strerror(errno));
			rte_errno = errno;
			return ret;
		}
		mod.qp_state = IBV_QPS_INIT;
		ret = mlx5_glue->modify_qp(qp, &mod,
					   (IBV_QP_STATE | IBV_QP_PORT));
		if (ret) {
			DRV_LOG(ERR, "Cannot change Tx QP state to INIT %s\n",
				strerror(errno));
			rte_errno = errno;
			return ret;
		}
		mod.qp_state = IBV_QPS_RTR;
		ret = mlx5_glue->modify_qp(qp, &mod, IBV_QP_STATE);
		if (ret) {
			DRV_LOG(ERR, "Cannot change Tx QP state to RTR %s\n",
				strerror(errno));
			rte_errno = errno;
			return ret;
		}
		mod.qp_state = IBV_QPS_RTS;
		ret = mlx5_glue->modify_qp(qp, &mod, IBV_QP_STATE);
		if (ret) {
			DRV_LOG(ERR, "Cannot change Tx QP state to RTS %s\n",
				strerror(errno));
			rte_errno = errno;
			return ret;
		}
	}
	return 0;
}

/**
 * Modify a Verbs queue state.
 *
 * @param dev
 *   Pointer to Ethernet device.
 * @param sm
 *   State modify request parameters.
 *
 * @return
 *   0 in case of success else non-zero value.
 */
static int
mlx5_queue_state_modify(struct rte_eth_dev *dev,
			struct mlx5_mp_arg_queue_state_modify *sm)
{
	int ret = 0;

	switch (rte_eal_process_type()) {
	case RTE_PROC_PRIMARY:
		ret = mlx5_queue_state_modify_primary(dev, sm);
		break;
	case RTE_PROC_SECONDARY:
		ret = mlx5_mp_req_queue_state_modify(dev, sm);
		break;
	default:
		break;
	}
	return ret;
}

/**
 * Handle a Rx error.
 * The function inserts the RQ state to reset when the first error CQE is
 * shown, then drains the CQ by the caller function loop. When the CQ is empty,
 * it moves the RQ state to ready and initializes the RQ.
 * Next CQE identification and error counting are in the caller responsibility.
 *
 * @param[in] rxq
 *   Pointer to RX queue structure.
 * @param[in] mbuf_prepare
 *   Whether to prepare mbufs for the RQ.
 *
 * @return
 *   -1 in case of recovery error, otherwise the CQE status.
 */
int
mlx5_rx_err_handle(struct mlx5_rxq_data *rxq, uint8_t mbuf_prepare)
{
	const uint16_t cqe_n = 1 << rxq->cqe_n;
	const uint16_t cqe_mask = cqe_n - 1;
	const unsigned int wqe_n = 1 << rxq->elts_n;
	struct mlx5_rxq_ctrl *rxq_ctrl =
			container_of(rxq, struct mlx5_rxq_ctrl, rxq);
	union {
		volatile struct mlx5_cqe *cqe;
		volatile struct mlx5_err_cqe *err_cqe;
	} u = {
		.cqe = &(*rxq->cqes)[rxq->cq_ci & cqe_mask],
	};
	struct mlx5_mp_arg_queue_state_modify sm;
	int ret;

	switch (rxq->err_state) {
	case MLX5_RXQ_ERR_STATE_NO_ERROR:
		rxq->err_state = MLX5_RXQ_ERR_STATE_NEED_RESET;
		/* Fall-through */
	case MLX5_RXQ_ERR_STATE_NEED_RESET:
		sm.is_wq = 1;
		sm.queue_id = rxq->idx;
		sm.state = IBV_WQS_RESET;
		if (mlx5_queue_state_modify(ETH_DEV(rxq_ctrl->priv), &sm))
			return -1;
		if (rxq_ctrl->dump_file_n <
		    rxq_ctrl->priv->config.max_dump_files_num) {
			MKSTR(err_str, "Unexpected CQE error syndrome "
			      "0x%02x CQN = %u RQN = %u wqe_counter = %u"
			      " rq_ci = %u cq_ci = %u", u.err_cqe->syndrome,
			      rxq->cqn, rxq_ctrl->wqn,
			      rte_be_to_cpu_16(u.err_cqe->wqe_counter),
			      rxq->rq_ci << rxq->sges_n, rxq->cq_ci);
			MKSTR(name, "dpdk_mlx5_port_%u_rxq_%u_%u",
			      rxq->port_id, rxq->idx, (uint32_t)rte_rdtsc());
			mlx5_dump_debug_information(name, NULL, err_str, 0);
			mlx5_dump_debug_information(name, "MLX5 Error CQ:",
						    (const void *)((uintptr_t)
								    rxq->cqes),
						    sizeof(*u.cqe) * cqe_n);
			mlx5_dump_debug_information(name, "MLX5 Error RQ:",
						    (const void *)((uintptr_t)
								    rxq->wqes),
						    16 * wqe_n);
			rxq_ctrl->dump_file_n++;
		}
		rxq->err_state = MLX5_RXQ_ERR_STATE_NEED_READY;
		/* Fall-through */
	case MLX5_RXQ_ERR_STATE_NEED_READY:
		ret = check_cqe(u.cqe, cqe_n, rxq->cq_ci);
		if (ret == MLX5_CQE_STATUS_HW_OWN) {
			rte_cio_wmb();
			*rxq->cq_db = rte_cpu_to_be_32(rxq->cq_ci);
			rte_cio_wmb();
			/*
			 * The RQ consumer index must be zeroed while moving
			 * from RESET state to RDY state.
			 */
			*rxq->rq_db = rte_cpu_to_be_32(0);
			rte_cio_wmb();
			sm.is_wq = 1;
			sm.queue_id = rxq->idx;
			sm.state = IBV_WQS_RDY;
			if (mlx5_queue_state_modify(ETH_DEV(rxq_ctrl->priv),
						    &sm))
				return -1;
			if (mbuf_prepare) {
				const uint16_t q_mask = wqe_n - 1;
				uint16_t elt_idx;
				struct rte_mbuf **elt;
				int i;
				unsigned int n = wqe_n - (rxq->rq_ci -
							  rxq->rq_pi);

				for (i = 0; i < (int)n; ++i) {
					elt_idx = (rxq->rq_ci + i) & q_mask;
					elt = &(*rxq->elts)[elt_idx];
					*elt = rte_mbuf_raw_alloc(rxq->mp);
					if (!*elt) {
						for (i--; i >= 0; --i) {
							elt_idx = (rxq->rq_ci +
								   i) & q_mask;
							elt = &(*rxq->elts)
								[elt_idx];
							rte_pktmbuf_free_seg
								(*elt);
						}
						return -1;
					}
				}
			}
			mlx5_rxq_initialize(rxq);
			rxq->err_state = MLX5_RXQ_ERR_STATE_NO_ERROR;
		}
		return ret;
	default:
		return -1;
	}
}

/**
 * Get size of the next packet for a given CQE. For compressed CQEs, the
 * consumer index is updated only once all packets of the current one have
 * been processed.
 *
 * @param rxq
 *   Pointer to RX queue.
 * @param cqe
 *   CQE to process.
 * @param[out] mcqe
 *   Store pointer to mini-CQE if compressed. Otherwise, the pointer is not
 *   written.
 *
 * @return
 *   0 in case of empty CQE, otherwise the packet size in bytes.
 */
static inline int
mlx5_rx_poll_len(struct mlx5_rxq_data *rxq, volatile struct mlx5_cqe *cqe,
		 uint16_t cqe_cnt, volatile struct mlx5_mini_cqe8 **mcqe)
{
	struct rxq_zip *zip = &rxq->zip;
	uint16_t cqe_n = cqe_cnt + 1;
	int len;
	uint16_t idx, end;

	do {
		len = 0;
		/* Process compressed data in the CQE and mini arrays. */
		if (zip->ai) {
			volatile struct mlx5_mini_cqe8 (*mc)[8] =
				(volatile struct mlx5_mini_cqe8 (*)[8])
				(uintptr_t)(&(*rxq->cqes)[zip->ca &
							  cqe_cnt].pkt_info);

			len = rte_be_to_cpu_32((*mc)[zip->ai & 7].byte_cnt);
			*mcqe = &(*mc)[zip->ai & 7];
			if ((++zip->ai & 7) == 0) {
				/* Invalidate consumed CQEs */
				idx = zip->ca;
				end = zip->na;
				while (idx != end) {
					(*rxq->cqes)[idx & cqe_cnt].op_own =
						MLX5_CQE_INVALIDATE;
					++idx;
				}
				/*
				 * Increment consumer index to skip the number
				 * of CQEs consumed. Hardware leaves holes in
				 * the CQ ring for software use.
				 */
				zip->ca = zip->na;
				zip->na += 8;
			}
			if (unlikely(rxq->zip.ai == rxq->zip.cqe_cnt)) {
				/* Invalidate the rest */
				idx = zip->ca;
				end = zip->cq_ci;

				while (idx != end) {
					(*rxq->cqes)[idx & cqe_cnt].op_own =
						MLX5_CQE_INVALIDATE;
					++idx;
				}
				rxq->cq_ci = zip->cq_ci;
				zip->ai = 0;
			}
		/*
		 * No compressed data, get next CQE and verify if it is
		 * compressed.
		 */
		} else {
			int ret;
			int8_t op_own;

			ret = check_cqe(cqe, cqe_n, rxq->cq_ci);
			if (unlikely(ret != MLX5_CQE_STATUS_SW_OWN)) {
				if (unlikely(ret == MLX5_CQE_STATUS_ERR ||
					     rxq->err_state)) {
					ret = mlx5_rx_err_handle(rxq, 0);
					if (ret == MLX5_CQE_STATUS_HW_OWN ||
					    ret == -1)
						return 0;
				} else {
					return 0;
				}
			}
			++rxq->cq_ci;
			op_own = cqe->op_own;
			if (MLX5_CQE_FORMAT(op_own) == MLX5_COMPRESSED) {
				volatile struct mlx5_mini_cqe8 (*mc)[8] =
					(volatile struct mlx5_mini_cqe8 (*)[8])
					(uintptr_t)(&(*rxq->cqes)
						[rxq->cq_ci &
						 cqe_cnt].pkt_info);

				/* Fix endianness. */
				zip->cqe_cnt = rte_be_to_cpu_32(cqe->byte_cnt);
				/*
				 * Current mini array position is the one
				 * returned by check_cqe64().
				 *
				 * If completion comprises several mini arrays,
				 * as a special case the second one is located
				 * 7 CQEs after the initial CQE instead of 8
				 * for subsequent ones.
				 */
				zip->ca = rxq->cq_ci;
				zip->na = zip->ca + 7;
				/* Compute the next non compressed CQE. */
				--rxq->cq_ci;
				zip->cq_ci = rxq->cq_ci + zip->cqe_cnt;
				/* Get packet size to return. */
				len = rte_be_to_cpu_32((*mc)[0].byte_cnt);
				*mcqe = &(*mc)[0];
				zip->ai = 1;
				/* Prefetch all to be invalidated */
				idx = zip->ca;
				end = zip->cq_ci;
				while (idx != end) {
					rte_prefetch0(&(*rxq->cqes)[(idx) &
								    cqe_cnt]);
					++idx;
				}
			} else {
				len = rte_be_to_cpu_32(cqe->byte_cnt);
			}
		}
		if (unlikely(rxq->err_state)) {
			cqe = &(*rxq->cqes)[rxq->cq_ci & cqe_cnt];
			++rxq->stats.idropped;
		} else {
			return len;
		}
	} while (1);
}

/**
 * Translate RX completion flags to offload flags.
 *
 * @param[in] cqe
 *   Pointer to CQE.
 *
 * @return
 *   Offload flags (ol_flags) for struct rte_mbuf.
 */
static inline uint32_t
rxq_cq_to_ol_flags(volatile struct mlx5_cqe *cqe)
{
	uint32_t ol_flags = 0;
	uint16_t flags = rte_be_to_cpu_16(cqe->hdr_type_etc);

	ol_flags =
		TRANSPOSE(flags,
			  MLX5_CQE_RX_L3_HDR_VALID,
			  PKT_RX_IP_CKSUM_GOOD) |
		TRANSPOSE(flags,
			  MLX5_CQE_RX_L4_HDR_VALID,
			  PKT_RX_L4_CKSUM_GOOD);
	return ol_flags;
}

/**
 * Fill in mbuf fields from RX completion flags.
 * Note that pkt->ol_flags should be initialized outside of this function.
 *
 * @param rxq
 *   Pointer to RX queue.
 * @param pkt
 *   mbuf to fill.
 * @param cqe
 *   CQE to process.
 * @param rss_hash_res
 *   Packet RSS Hash result.
 */
static inline void
rxq_cq_to_mbuf(struct mlx5_rxq_data *rxq, struct rte_mbuf *pkt,
	       volatile struct mlx5_cqe *cqe, uint32_t rss_hash_res)
{
	/* Update packet information. */
	pkt->packet_type = rxq_cq_to_pkt_type(rxq, cqe);
	if (rss_hash_res && rxq->rss_hash) {
		pkt->hash.rss = rss_hash_res;
		pkt->ol_flags |= PKT_RX_RSS_HASH;
	}
	if (rxq->mark && MLX5_FLOW_MARK_IS_VALID(cqe->sop_drop_qpn)) {
		pkt->ol_flags |= PKT_RX_FDIR;
		if (cqe->sop_drop_qpn !=
		    rte_cpu_to_be_32(MLX5_FLOW_MARK_DEFAULT)) {
			uint32_t mark = cqe->sop_drop_qpn;

			pkt->ol_flags |= PKT_RX_FDIR_ID;
			pkt->hash.fdir.hi = mlx5_flow_mark_get(mark);
		}
	}
	if (rxq->csum)
		pkt->ol_flags |= rxq_cq_to_ol_flags(cqe);
	if (rxq->vlan_strip &&
	    (cqe->hdr_type_etc & rte_cpu_to_be_16(MLX5_CQE_VLAN_STRIPPED))) {
		pkt->ol_flags |= PKT_RX_VLAN | PKT_RX_VLAN_STRIPPED;
		pkt->vlan_tci = rte_be_to_cpu_16(cqe->vlan_info);
	}
	if (rxq->hw_timestamp) {
		pkt->timestamp = rte_be_to_cpu_64(cqe->timestamp);
		pkt->ol_flags |= PKT_RX_TIMESTAMP;
	}
}

/**
 * DPDK callback for RX.
 *
 * @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(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
{
	struct mlx5_rxq_data *rxq = dpdk_rxq;
	const unsigned int wqe_cnt = (1 << rxq->elts_n) - 1;
	const unsigned int cqe_cnt = (1 << rxq->cqe_n) - 1;
	const unsigned int sges_n = rxq->sges_n;
	struct rte_mbuf *pkt = NULL;
	struct rte_mbuf *seg = NULL;
	volatile struct mlx5_cqe *cqe =
		&(*rxq->cqes)[rxq->cq_ci & cqe_cnt];
	unsigned int i = 0;
	unsigned int rq_ci = rxq->rq_ci << sges_n;
	int len = 0; /* keep its value across iterations. */

	while (pkts_n) {
		unsigned int idx = rq_ci & wqe_cnt;
		volatile struct mlx5_wqe_data_seg *wqe =
			&((volatile struct mlx5_wqe_data_seg *)rxq->wqes)[idx];
		struct rte_mbuf *rep = (*rxq->elts)[idx];
		volatile struct mlx5_mini_cqe8 *mcqe = NULL;
		uint32_t rss_hash_res;

		if (pkt)
			NEXT(seg) = rep;
		seg = rep;
		rte_prefetch0(seg);
		rte_prefetch0(cqe);
		rte_prefetch0(wqe);
		rep = rte_mbuf_raw_alloc(rxq->mp);
		if (unlikely(rep == NULL)) {
			++rxq->stats.rx_nombuf;
			if (!pkt) {
				/*
				 * no buffers before we even started,
				 * bail out silently.
				 */
				break;
			}
			while (pkt != seg) {
				assert(pkt != (*rxq->elts)[idx]);
				rep = NEXT(pkt);
				NEXT(pkt) = NULL;
				NB_SEGS(pkt) = 1;
				rte_mbuf_raw_free(pkt);
				pkt = rep;
			}
			break;
		}
		if (!pkt) {
			cqe = &(*rxq->cqes)[rxq->cq_ci & cqe_cnt];
			len = mlx5_rx_poll_len(rxq, cqe, cqe_cnt, &mcqe);
			if (!len) {
				rte_mbuf_raw_free(rep);
				break;
			}
			pkt = seg;
			assert(len >= (rxq->crc_present << 2));
			pkt->ol_flags = 0;
			/* If compressed, take hash result from mini-CQE. */
			rss_hash_res = rte_be_to_cpu_32(mcqe == NULL ?
							cqe->rx_hash_res :
							mcqe->rx_hash_result);
			rxq_cq_to_mbuf(rxq, pkt, cqe, rss_hash_res);
			if (rxq->crc_present)
				len -= RTE_ETHER_CRC_LEN;
			PKT_LEN(pkt) = len;
			if (cqe->lro_num_seg > 1) {
				mlx5_lro_update_hdr
					(rte_pktmbuf_mtod(pkt, uint8_t *), cqe,
					 len);
				pkt->ol_flags |= PKT_RX_LRO;
				pkt->tso_segsz = len / cqe->lro_num_seg;
			}
		}
		DATA_LEN(rep) = DATA_LEN(seg);
		PKT_LEN(rep) = PKT_LEN(seg);
		SET_DATA_OFF(rep, DATA_OFF(seg));
		PORT(rep) = PORT(seg);
		(*rxq->elts)[idx] = rep;
		/*
		 * Fill NIC descriptor with the new buffer.  The lkey and size
		 * of the buffers are already known, only the buffer address
		 * changes.
		 */
		wqe->addr = rte_cpu_to_be_64(rte_pktmbuf_mtod(rep, uintptr_t));
		/* If there's only one MR, no need to replace LKey in WQE. */
		if (unlikely(mlx5_mr_btree_len(&rxq->mr_ctrl.cache_bh) > 1))
			wqe->lkey = mlx5_rx_mb2mr(rxq, rep);
		if (len > DATA_LEN(seg)) {
			len -= DATA_LEN(seg);
			++NB_SEGS(pkt);
			++rq_ci;
			continue;
		}
		DATA_LEN(seg) = len;
#ifdef MLX5_PMD_SOFT_COUNTERS
		/* Increment bytes counter. */
		rxq->stats.ibytes += PKT_LEN(pkt);
#endif
		/* Return packet. */
		*(pkts++) = pkt;
		pkt = NULL;
		--pkts_n;
		++i;
		/* Align consumer index to the next stride. */
		rq_ci >>= sges_n;
		++rq_ci;
		rq_ci <<= sges_n;
	}
	if (unlikely((i == 0) && ((rq_ci >> sges_n) == rxq->rq_ci)))
		return 0;
	/* Update the consumer index. */
	rxq->rq_ci = rq_ci >> sges_n;
	rte_cio_wmb();
	*rxq->cq_db = rte_cpu_to_be_32(rxq->cq_ci);
	rte_cio_wmb();
	*rxq->rq_db = rte_cpu_to_be_32(rxq->rq_ci);
#ifdef MLX5_PMD_SOFT_COUNTERS
	/* Increment packets counter. */
	rxq->stats.ipackets += i;
#endif
	return i;
}

/**
 * Update LRO packet TCP header.
 * The HW LRO feature doesn't update the TCP header after coalescing the
 * TCP segments but supplies information in CQE to fill it by SW.
 *
 * @param tcp
 *   Pointer to the TCP header.
 * @param cqe
 *   Pointer to the completion entry..
 * @param phcsum
 *   The L3 pseudo-header checksum.
 */
static inline void
mlx5_lro_update_tcp_hdr(struct rte_tcp_hdr *restrict tcp,
			volatile struct mlx5_cqe *restrict cqe,
			uint32_t phcsum)
{
	uint8_t l4_type = (rte_be_to_cpu_16(cqe->hdr_type_etc) &
			   MLX5_CQE_L4_TYPE_MASK) >> MLX5_CQE_L4_TYPE_SHIFT;
	/*
	 * The HW calculates only the TCP payload checksum, need to complete
	 * the TCP header checksum and the L3 pseudo-header checksum.
	 */
	uint32_t csum = phcsum + cqe->csum;

	if (l4_type == MLX5_L4_HDR_TYPE_TCP_EMPTY_ACK ||
	    l4_type == MLX5_L4_HDR_TYPE_TCP_WITH_ACL) {
		tcp->tcp_flags |= RTE_TCP_ACK_FLAG;
		tcp->recv_ack = cqe->lro_ack_seq_num;
		tcp->rx_win = cqe->lro_tcp_win;
	}
	if (cqe->lro_tcppsh_abort_dupack & MLX5_CQE_LRO_PUSH_MASK)
		tcp->tcp_flags |= RTE_TCP_PSH_FLAG;
	tcp->cksum = 0;
	csum += rte_raw_cksum(tcp, (tcp->data_off & 0xF) * 4);
	csum = ((csum & 0xffff0000) >> 16) + (csum & 0xffff);
	csum = (~csum) & 0xffff;
	if (csum == 0)
		csum = 0xffff;
	tcp->cksum = csum;
}

/**
 * Update LRO packet headers.
 * The HW LRO feature doesn't update the L3/TCP headers after coalescing the
 * TCP segments but supply information in CQE to fill it by SW.
 *
 * @param padd
 *   The packet address.
 * @param cqe
 *   Pointer to the completion entry..
 * @param len
 *   The packet length.
 */
static inline void
mlx5_lro_update_hdr(uint8_t *restrict padd,
		    volatile struct mlx5_cqe *restrict cqe,
		    uint32_t len)
{
	union {
		struct rte_ether_hdr *eth;
		struct rte_vlan_hdr *vlan;
		struct rte_ipv4_hdr *ipv4;
		struct rte_ipv6_hdr *ipv6;
		struct rte_tcp_hdr *tcp;
		uint8_t *hdr;
	} h = {
			.hdr = padd,
	};
	uint16_t proto = h.eth->ether_type;
	uint32_t phcsum;

	h.eth++;
	while (proto == RTE_BE16(RTE_ETHER_TYPE_VLAN) ||
	       proto == RTE_BE16(RTE_ETHER_TYPE_QINQ)) {
		proto = h.vlan->eth_proto;
		h.vlan++;
	}
	if (proto == RTE_BE16(RTE_ETHER_TYPE_IPV4)) {
		h.ipv4->time_to_live = cqe->lro_min_ttl;
		h.ipv4->total_length = rte_cpu_to_be_16(len - (h.hdr - padd));
		h.ipv4->hdr_checksum = 0;
		h.ipv4->hdr_checksum = rte_ipv4_cksum(h.ipv4);
		phcsum = rte_ipv4_phdr_cksum(h.ipv4, 0);
		h.ipv4++;
	} else {
		h.ipv6->hop_limits = cqe->lro_min_ttl;
		h.ipv6->payload_len = rte_cpu_to_be_16(len - (h.hdr - padd) -
						       sizeof(*h.ipv6));
		phcsum = rte_ipv6_phdr_cksum(h.ipv6, 0);
		h.ipv6++;
	}
	mlx5_lro_update_tcp_hdr(h.tcp, cqe, phcsum);
}

void
mlx5_mprq_buf_free_cb(void *addr __rte_unused, void *opaque)
{
	struct mlx5_mprq_buf *buf = opaque;

	if (rte_atomic16_read(&buf->refcnt) == 1) {
		rte_mempool_put(buf->mp, buf);
	} else if (rte_atomic16_add_return(&buf->refcnt, -1) == 0) {
		rte_atomic16_set(&buf->refcnt, 1);
		rte_mempool_put(buf->mp, buf);
	}
}

void
mlx5_mprq_buf_free(struct mlx5_mprq_buf *buf)
{
	mlx5_mprq_buf_free_cb(NULL, buf);
}

static inline void
mprq_buf_replace(struct mlx5_rxq_data *rxq, uint16_t rq_idx,
		 const unsigned int strd_n)
{
	struct mlx5_mprq_buf *rep = rxq->mprq_repl;
	volatile struct mlx5_wqe_data_seg *wqe =
		&((volatile struct mlx5_wqe_mprq *)rxq->wqes)[rq_idx].dseg;
	void *addr;

	assert(rep != NULL);
	/* Replace MPRQ buf. */
	(*rxq->mprq_bufs)[rq_idx] = rep;
	/* Replace WQE. */
	addr = mlx5_mprq_buf_addr(rep, strd_n);
	wqe->addr = rte_cpu_to_be_64((uintptr_t)addr);
	/* If there's only one MR, no need to replace LKey in WQE. */
	if (unlikely(mlx5_mr_btree_len(&rxq->mr_ctrl.cache_bh) > 1))
		wqe->lkey = mlx5_rx_addr2mr(rxq, (uintptr_t)addr);
	/* Stash a mbuf for next replacement. */
	if (likely(!rte_mempool_get(rxq->mprq_mp, (void **)&rep)))
		rxq->mprq_repl = rep;
	else
		rxq->mprq_repl = NULL;
}

/**
 * DPDK callback for RX with Multi-Packet RQ support.
 *
 * @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_mprq(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
{
	struct mlx5_rxq_data *rxq = dpdk_rxq;
	const unsigned int strd_n = 1 << rxq->strd_num_n;
	const unsigned int strd_sz = 1 << rxq->strd_sz_n;
	const unsigned int strd_shift =
		MLX5_MPRQ_STRIDE_SHIFT_BYTE * rxq->strd_shift_en;
	const unsigned int cq_mask = (1 << rxq->cqe_n) - 1;
	const unsigned int wq_mask = (1 << rxq->elts_n) - 1;
	volatile struct mlx5_cqe *cqe = &(*rxq->cqes)[rxq->cq_ci & cq_mask];
	unsigned int i = 0;
	uint32_t rq_ci = rxq->rq_ci;
	uint16_t consumed_strd = rxq->consumed_strd;
	uint16_t headroom_sz = rxq->strd_headroom_en * RTE_PKTMBUF_HEADROOM;
	struct mlx5_mprq_buf *buf = (*rxq->mprq_bufs)[rq_ci & wq_mask];

	while (i < pkts_n) {
		struct rte_mbuf *pkt;
		void *addr;
		int ret;
		unsigned int len;
		uint16_t strd_cnt;
		uint16_t strd_idx;
		uint32_t offset;
		uint32_t byte_cnt;
		volatile struct mlx5_mini_cqe8 *mcqe = NULL;
		uint32_t rss_hash_res = 0;
		uint8_t lro_num_seg;

		if (consumed_strd == strd_n) {
			/* Replace WQE only if the buffer is still in use. */
			if (rte_atomic16_read(&buf->refcnt) > 1) {
				mprq_buf_replace(rxq, rq_ci & wq_mask, strd_n);
				/* Release the old buffer. */
				mlx5_mprq_buf_free(buf);
			} else if (unlikely(rxq->mprq_repl == NULL)) {
				struct mlx5_mprq_buf *rep;

				/*
				 * Currently, the MPRQ mempool is out of buffer
				 * and doing memcpy regardless of the size of Rx
				 * packet. Retry allocation to get back to
				 * normal.
				 */
				if (!rte_mempool_get(rxq->mprq_mp,
						     (void **)&rep))
					rxq->mprq_repl = rep;
			}
			/* Advance to the next WQE. */
			consumed_strd = 0;
			++rq_ci;
			buf = (*rxq->mprq_bufs)[rq_ci & wq_mask];
		}
		cqe = &(*rxq->cqes)[rxq->cq_ci & cq_mask];
		ret = mlx5_rx_poll_len(rxq, cqe, cq_mask, &mcqe);
		if (!ret)
			break;
		byte_cnt = ret;
		strd_cnt = (byte_cnt & MLX5_MPRQ_STRIDE_NUM_MASK) >>
			   MLX5_MPRQ_STRIDE_NUM_SHIFT;
		assert(strd_cnt);
		consumed_strd += strd_cnt;
		if (byte_cnt & MLX5_MPRQ_FILLER_MASK)
			continue;
		if (mcqe == NULL) {
			rss_hash_res = rte_be_to_cpu_32(cqe->rx_hash_res);
			strd_idx = rte_be_to_cpu_16(cqe->wqe_counter);
		} else {
			/* mini-CQE for MPRQ doesn't have hash result. */
			strd_idx = rte_be_to_cpu_16(mcqe->stride_idx);
		}
		assert(strd_idx < strd_n);
		assert(!((rte_be_to_cpu_16(cqe->wqe_id) ^ rq_ci) & wq_mask));
		lro_num_seg = cqe->lro_num_seg;
		/*
		 * Currently configured to receive a packet per a stride. But if
		 * MTU is adjusted through kernel interface, device could
		 * consume multiple strides without raising an error. In this
		 * case, the packet should be dropped because it is bigger than
		 * the max_rx_pkt_len.
		 */
		if (unlikely(!lro_num_seg && strd_cnt > 1)) {
			++rxq->stats.idropped;
			continue;
		}
		pkt = rte_pktmbuf_alloc(rxq->mp);
		if (unlikely(pkt == NULL)) {
			++rxq->stats.rx_nombuf;
			break;
		}
		len = (byte_cnt & MLX5_MPRQ_LEN_MASK) >> MLX5_MPRQ_LEN_SHIFT;
		assert((int)len >= (rxq->crc_present << 2));
		if (rxq->crc_present)
			len -= RTE_ETHER_CRC_LEN;
		offset = strd_idx * strd_sz + strd_shift;
		addr = RTE_PTR_ADD(mlx5_mprq_buf_addr(buf, strd_n), offset);
		/*
		 * Memcpy packets to the target mbuf if:
		 * - The size of packet is smaller than mprq_max_memcpy_len.
		 * - Out of buffer in the Mempool for Multi-Packet RQ.
		 */
		if (len <= rxq->mprq_max_memcpy_len || rxq->mprq_repl == NULL) {
			/*
			 * When memcpy'ing packet due to out-of-buffer, the
			 * packet must be smaller than the target mbuf.
			 */
			if (unlikely(rte_pktmbuf_tailroom(pkt) < len)) {
				rte_pktmbuf_free_seg(pkt);
				++rxq->stats.idropped;
				continue;
			}
			rte_memcpy(rte_pktmbuf_mtod(pkt, void *), addr, len);
			DATA_LEN(pkt) = len;
		} else {
			rte_iova_t buf_iova;
			struct rte_mbuf_ext_shared_info *shinfo;
			uint16_t buf_len = strd_cnt * strd_sz;
			void *buf_addr;

			/* Increment the refcnt of the whole chunk. */
			rte_atomic16_add_return(&buf->refcnt, 1);
			assert((uint16_t)rte_atomic16_read(&buf->refcnt) <=
			       strd_n + 1);
			buf_addr = RTE_PTR_SUB(addr, headroom_sz);
			/*
			 * MLX5 device doesn't use iova but it is necessary in a
			 * case where the Rx packet is transmitted via a
			 * different PMD.
			 */
			buf_iova = rte_mempool_virt2iova(buf) +
				   RTE_PTR_DIFF(buf_addr, buf);
			shinfo = &buf->shinfos[strd_idx];
			rte_mbuf_ext_refcnt_set(shinfo, 1);
			/*
			 * EXT_ATTACHED_MBUF will be set to pkt->ol_flags when
			 * attaching the stride to mbuf and more offload flags
			 * will be added below by calling rxq_cq_to_mbuf().
			 * Other fields will be overwritten.
			 */
			rte_pktmbuf_attach_extbuf(pkt, buf_addr, buf_iova,
						  buf_len, shinfo);
			/* Set mbuf head-room. */
			pkt->data_off = headroom_sz;
			assert(pkt->ol_flags == EXT_ATTACHED_MBUF);
			/*
			 * Prevent potential overflow due to MTU change through
			 * kernel interface.
			 */
			if (unlikely(rte_pktmbuf_tailroom(pkt) < len)) {
				rte_pktmbuf_free_seg(pkt);
				++rxq->stats.idropped;
				continue;
			}
			DATA_LEN(pkt) = len;
			/*
			 * LRO packet may consume all the stride memory, in this
			 * case packet head-room space is not guaranteed so must
			 * to add an empty mbuf for the head-room.
			 */
			if (!rxq->strd_headroom_en) {
				struct rte_mbuf *headroom_mbuf =
						rte_pktmbuf_alloc(rxq->mp);

				if (unlikely(headroom_mbuf == NULL)) {
					rte_pktmbuf_free_seg(pkt);
					++rxq->stats.rx_nombuf;
					break;
				}
				PORT(pkt) = rxq->port_id;
				NEXT(headroom_mbuf) = pkt;
				pkt = headroom_mbuf;
				NB_SEGS(pkt) = 2;
			}
		}
		rxq_cq_to_mbuf(rxq, pkt, cqe, rss_hash_res);
		if (lro_num_seg > 1) {
			mlx5_lro_update_hdr(addr, cqe, len);
			pkt->ol_flags |= PKT_RX_LRO;
			pkt->tso_segsz = strd_sz;
		}
		PKT_LEN(pkt) = len;
		PORT(pkt) = rxq->port_id;
#ifdef MLX5_PMD_SOFT_COUNTERS
		/* Increment bytes counter. */
		rxq->stats.ibytes += PKT_LEN(pkt);
#endif
		/* Return packet. */
		*(pkts++) = pkt;
		++i;
	}
	/* Update the consumer indexes. */
	rxq->consumed_strd = consumed_strd;
	rte_cio_wmb();
	*rxq->cq_db = rte_cpu_to_be_32(rxq->cq_ci);
	if (rq_ci != rxq->rq_ci) {
		rxq->rq_ci = rq_ci;
		rte_cio_wmb();
		*rxq->rq_db = rte_cpu_to_be_32(rxq->rq_ci);
	}
#ifdef MLX5_PMD_SOFT_COUNTERS
	/* Increment packets counter. */
	rxq->stats.ipackets += i;
#endif
	return i;
}

/**
 * Dummy DPDK callback for TX.
 *
 * This function is used to temporarily replace the real callback during
 * unsafe control operations on the queue, or in case of error.
 *
 * @param dpdk_txq
 *   Generic pointer to TX queue structure.
 * @param[in] pkts
 *   Packets to transmit.
 * @param pkts_n
 *   Number of packets in array.
 *
 * @return
 *   Number of packets successfully transmitted (<= pkts_n).
 */
uint16_t
removed_tx_burst(void *dpdk_txq __rte_unused,
		 struct rte_mbuf **pkts __rte_unused,
		 uint16_t pkts_n __rte_unused)
{
	rte_mb();
	return 0;
}

/**
 * Dummy DPDK callback for RX.
 *
 * This function is used to temporarily replace the real callback during
 * unsafe control operations on the queue, or in case of error.
 *
 * @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
removed_rx_burst(void *dpdk_txq __rte_unused,
		 struct rte_mbuf **pkts __rte_unused,
		 uint16_t pkts_n __rte_unused)
{
	rte_mb();
	return 0;
}

/*
 * Vectorized Rx/Tx routines are not compiled in when required vector
 * instructions are not supported on a target architecture. The following null
 * stubs are needed for linkage when those are not included outside of this file
 * (e.g.  mlx5_rxtx_vec_sse.c for x86).
 */

__rte_weak uint16_t
mlx5_rx_burst_vec(void *dpdk_txq __rte_unused,
		  struct rte_mbuf **pkts __rte_unused,
		  uint16_t pkts_n __rte_unused)
{
	return 0;
}

__rte_weak int
mlx5_rxq_check_vec_support(struct mlx5_rxq_data *rxq __rte_unused)
{
	return -ENOTSUP;
}

__rte_weak int
mlx5_check_vec_rx_support(struct rte_eth_dev *dev __rte_unused)
{
	return -ENOTSUP;
}

/**
 * Free the mbufs from the linear array of pointers.
 *
 * @param pkts
 *   Pointer to array of packets to be free.
 * @param pkts_n
 *   Number of packets to be freed.
 * @param olx
 *   Configured Tx offloads mask. It is fully defined at
 *   compile time and may be used for optimization.
 */
static __rte_always_inline void
mlx5_tx_free_mbuf(struct rte_mbuf **restrict pkts,
		  unsigned int pkts_n,
		  unsigned int olx __rte_unused)
{
	struct rte_mempool *pool = NULL;
	struct rte_mbuf **p_free = NULL;
	struct rte_mbuf *mbuf;
	unsigned int n_free = 0;

	/*
	 * The implemented algorithm eliminates
	 * copying pointers to temporary array
	 * for rte_mempool_put_bulk() calls.
	 */
	assert(pkts);
	assert(pkts_n);
	for (;;) {
		for (;;) {
			/*
			 * Decrement mbuf reference counter, detach
			 * indirect and external buffers if needed.
			 */
			mbuf = rte_pktmbuf_prefree_seg(*pkts);
			if (likely(mbuf != NULL)) {
				assert(mbuf == *pkts);
				if (likely(n_free != 0)) {
					if (unlikely(pool != mbuf->pool))
						/* From different pool. */
						break;
				} else {
					/* Start new scan array. */
					pool = mbuf->pool;
					p_free = pkts;
				}
				++n_free;
				++pkts;
				--pkts_n;
				if (unlikely(pkts_n == 0)) {
					mbuf = NULL;
					break;
				}
			} else {
				/*
				 * This happens if mbuf is still referenced.
				 * We can't put it back to the pool, skip.
				 */
				++pkts;
				--pkts_n;
				if (unlikely(n_free != 0))
					/* There is some array to free.*/
					break;
				if (unlikely(pkts_n == 0))
					/* Last mbuf, nothing to free. */
					return;
			}
		}
		for (;;) {
			/*
			 * This loop is implemented to avoid multiple
			 * inlining of rte_mempool_put_bulk().
			 */
			assert(pool);
			assert(p_free);
			assert(n_free);
			/*
			 * Free the array of pre-freed mbufs
			 * belonging to the same memory pool.
			 */
			rte_mempool_put_bulk(pool, (void *)p_free, n_free);
			if (unlikely(mbuf != NULL)) {
				/* There is the request to start new scan. */
				pool = mbuf->pool;
				p_free = pkts++;
				n_free = 1;
				--pkts_n;
				if (likely(pkts_n != 0))
					break;
				/*
				 * This is the last mbuf to be freed.
				 * Do one more loop iteration to complete.
				 * This is rare case of the last unique mbuf.
				 */
				mbuf = NULL;
				continue;
			}
			if (likely(pkts_n == 0))
				return;
			n_free = 0;
			break;
		}
	}
}

/**
 * Free the mbuf from the elts ring buffer till new tail.
 *
 * @param txq
 *   Pointer to Tx queue structure.
 * @param tail
 *   Index in elts to free up to, becomes new elts tail.
 * @param olx
 *   Configured Tx offloads mask. It is fully defined at
 *   compile time and may be used for optimization.
 */
static __rte_always_inline void
mlx5_tx_free_elts(struct mlx5_txq_data *restrict txq,
		  uint16_t tail,
		  unsigned int olx __rte_unused)
{
	uint16_t n_elts = tail - txq->elts_tail;

	assert(n_elts);
	assert(n_elts <= txq->elts_s);
	/*
	 * Implement a loop to support ring buffer wraparound
	 * with single inlining of mlx5_tx_free_mbuf().
	 */
	do {
		unsigned int part;

		part = txq->elts_s - (txq->elts_tail & txq->elts_m);
		part = RTE_MIN(part, n_elts);
		assert(part);
		assert(part <= txq->elts_s);
		mlx5_tx_free_mbuf(&txq->elts[txq->elts_tail & txq->elts_m],
				  part, olx);
		txq->elts_tail += part;
		n_elts -= part;
	} while (n_elts);
}

/**
 * Store the mbuf being sent into elts ring buffer.
 * On Tx completion these mbufs will be freed.
 *
 * @param txq
 *   Pointer to Tx queue structure.
 * @param pkts
 *   Pointer to array of packets to be stored.
 * @param pkts_n
 *   Number of packets to be stored.
 * @param olx
 *   Configured Tx offloads mask. It is fully defined at
 *   compile time and may be used for optimization.
 */
static __rte_always_inline void
mlx5_tx_copy_elts(struct mlx5_txq_data *restrict txq,
		  struct rte_mbuf **restrict pkts,
		  unsigned int pkts_n,
		  unsigned int olx __rte_unused)
{
	unsigned int part;
	struct rte_mbuf **elts = (struct rte_mbuf **)txq->elts;

	assert(pkts);
	assert(pkts_n);
	part = txq->elts_s - (txq->elts_head & txq->elts_m);
	assert(part);
	assert(part <= txq->elts_s);
	/* This code is a good candidate for vectorizing with SIMD. */
	rte_memcpy((void *)(elts + (txq->elts_head & txq->elts_m)),
		   (void *)pkts,
		   RTE_MIN(part, pkts_n) * sizeof(struct rte_mbuf *));
	txq->elts_head += pkts_n;
	if (unlikely(part < pkts_n))
		/* The copy is wrapping around the elts array. */
		rte_memcpy((void *)elts, (void *)(pkts + part),
			   (pkts_n - part) * sizeof(struct rte_mbuf *));
}

/**
 * Update completion queue consuming index via doorbell
 * and flush the completed data buffers.
 *
 * @param txq
 *   Pointer to TX queue structure.
 * @param valid CQE pointer
 *   if not NULL update txq->wqe_pi and flush the buffers
 * @param itail
 *   if not negative - flush the buffers till this index.
 * @param olx
 *   Configured Tx offloads mask. It is fully defined at
 *   compile time and may be used for optimization.
 */
static __rte_always_inline void
mlx5_tx_comp_flush(struct mlx5_txq_data *restrict txq,
		   volatile struct mlx5_cqe *last_cqe,
		   int itail,
		   unsigned int olx __rte_unused)
{
	uint16_t tail;

	if (likely(last_cqe != NULL)) {
		txq->wqe_pi = rte_be_to_cpu_16(last_cqe->wqe_counter);
		tail = ((volatile struct mlx5_wqe_cseg *)
			(txq->wqes + (txq->wqe_pi & txq->wqe_m)))->misc;
	} else if (itail >= 0) {
		tail = (uint16_t)itail;
	} else {
		return;
	}
	rte_compiler_barrier();
	*txq->cq_db = rte_cpu_to_be_32(txq->cq_ci);
	if (likely(tail != txq->elts_tail)) {
		mlx5_tx_free_elts(txq, tail, olx);
		assert(tail == txq->elts_tail);
	}
}

/**
 * Manage TX completions. This routine checks the CQ for
 * arrived CQEs, deduces the last accomplished WQE in SQ,
 * updates SQ producing index and frees all completed mbufs.
 *
 * @param txq
 *   Pointer to TX queue structure.
 * @param olx
 *   Configured Tx offloads mask. It is fully defined at
 *   compile time and may be used for optimization.
 *
 * NOTE: not inlined intentionally, it makes tx_burst
 * routine smaller, simple and faster - from experiments.
 */
static void
mlx5_tx_handle_completion(struct mlx5_txq_data *restrict txq,
			  unsigned int olx __rte_unused)
{
	unsigned int count = MLX5_TX_COMP_MAX_CQE;
	volatile struct mlx5_cqe *last_cqe = NULL;
	int ret;

	static_assert(MLX5_CQE_STATUS_HW_OWN < 0, "Must be negative value");
	static_assert(MLX5_CQE_STATUS_SW_OWN < 0, "Must be negative value");
	do {
		volatile struct mlx5_cqe *cqe;

		cqe = &txq->cqes[txq->cq_ci & txq->cqe_m];
		ret = check_cqe(cqe, txq->cqe_s, txq->cq_ci);
		if (unlikely(ret != MLX5_CQE_STATUS_SW_OWN)) {
			if (likely(ret != MLX5_CQE_STATUS_ERR)) {
				/* No new CQEs in completion queue. */
				assert(ret == MLX5_CQE_STATUS_HW_OWN);
				break;
			}
			/*
			 * Some error occurred, try to restart.
			 * We have no barrier after WQE related Doorbell
			 * written, make sure all writes are completed
			 * here, before we might perform SQ reset.
			 */
			rte_wmb();
			ret = mlx5_tx_error_cqe_handle
				(txq, (volatile struct mlx5_err_cqe *)cqe);
			/*
			 * Flush buffers, update consuming index
			 * if recovery succeeded. Otherwise
			 * just try to recover later.
			 */
			last_cqe = NULL;
			break;
		}
		/* Normal transmit completion. */
		++txq->cq_ci;
		last_cqe = cqe;
#ifndef NDEBUG
		if (txq->cq_pi)
			--txq->cq_pi;
#endif
	/*
	 * We have to restrict the amount of processed CQEs
	 * in one tx_burst routine call. The CQ may be large
	 * and many CQEs may be updated by the NIC in one
	 * transaction. Buffers freeing is time consuming,
	 * multiple iterations may introduce significant
	 * latency.
	 */
	} while (--count);
	mlx5_tx_comp_flush(txq, last_cqe, ret, olx);
}

/**
 * Check if the completion request flag should be set in the last WQE.
 * Both pushed mbufs and WQEs are monitored and the completion request
 * flag is set if any of thresholds is reached.
 *
 * @param txq
 *   Pointer to TX queue structure.
 * @param loc
 *   Pointer to burst routine local context.
 * @param multi,
 *   Routine is called from multi-segment sending loop,
 *   do not correct the elts_head according to the pkts_copy.
 * @param olx
 *   Configured Tx offloads mask. It is fully defined at
 *   compile time and may be used for optimization.
 */
static __rte_always_inline void
mlx5_tx_request_completion(struct mlx5_txq_data *restrict txq,
			   struct mlx5_txq_local *restrict loc,
			   bool multi,
			   unsigned int olx)
{
	uint16_t head = txq->elts_head;
	unsigned int part;

	part = (MLX5_TXOFF_CONFIG(INLINE) || multi) ?
	       0 : loc->pkts_sent - loc->pkts_copy;
	head += part;
	if ((uint16_t)(head - txq->elts_comp) >= MLX5_TX_COMP_THRESH ||
	     (MLX5_TXOFF_CONFIG(INLINE) &&
	     (uint16_t)(txq->wqe_ci - txq->wqe_comp) >= txq->wqe_thres)) {
		volatile struct mlx5_wqe *last = loc->wqe_last;

		txq->elts_comp = head;
		if (MLX5_TXOFF_CONFIG(INLINE))
			txq->wqe_comp = txq->wqe_ci;
		/* Request unconditional completion on last WQE. */
		last->cseg.flags = RTE_BE32(MLX5_COMP_ALWAYS <<
					    MLX5_COMP_MODE_OFFSET);
		/* Save elts_head in unused "immediate" field of WQE. */
		last->cseg.misc = head;
		/*
		 * A CQE slot must always be available. Count the
		 * issued CEQ "always" request instead of production
		 * index due to here can be CQE with errors and
		 * difference with ci may become inconsistent.
		 */
		assert(txq->cqe_s > ++txq->cq_pi);
	}
}

/**
 * DPDK callback to check the status of a tx descriptor.
 *
 * @param tx_queue
 *   The tx queue.
 * @param[in] offset
 *   The index of the descriptor in the ring.
 *
 * @return
 *   The status of the tx descriptor.
 */
int
mlx5_tx_descriptor_status(void *tx_queue, uint16_t offset)
{
	struct mlx5_txq_data *restrict txq = tx_queue;
	uint16_t used;

	mlx5_tx_handle_completion(txq, 0);
	used = txq->elts_head - txq->elts_tail;
	if (offset < used)
		return RTE_ETH_TX_DESC_FULL;
	return RTE_ETH_TX_DESC_DONE;
}

/**
 * Build the Control Segment with specified opcode:
 * - MLX5_OPCODE_SEND
 * - MLX5_OPCODE_ENHANCED_MPSW
 * - MLX5_OPCODE_TSO
 *
 * @param txq
 *   Pointer to TX queue structure.
 * @param loc
 *   Pointer to burst routine local context.
 * @param wqe
 *   Pointer to WQE to fill with built Control Segment.
 * @param ds
 *   Supposed length of WQE in segments.
 * @param opcode
 *   SQ WQE opcode to put into Control Segment.
 * @param olx
 *   Configured Tx offloads mask. It is fully defined at
 *   compile time and may be used for optimization.
 */
static __rte_always_inline void
mlx5_tx_cseg_init(struct mlx5_txq_data *restrict txq,
		  struct mlx5_txq_local *restrict loc __rte_unused,
		  struct mlx5_wqe *restrict wqe,
		  unsigned int ds,
		  unsigned int opcode,
		  unsigned int olx __rte_unused)
{
	struct mlx5_wqe_cseg *restrict cs = &wqe->cseg;

	cs->opcode = rte_cpu_to_be_32((txq->wqe_ci << 8) | opcode);
	cs->sq_ds = rte_cpu_to_be_32(txq->qp_num_8s | ds);
	cs->flags = RTE_BE32(MLX5_COMP_ONLY_FIRST_ERR <<
			     MLX5_COMP_MODE_OFFSET);
	cs->misc = RTE_BE32(0);
}

/**
 * Build the Ethernet Segment without inlined data.
 * Supports Software Parser, Checksums and VLAN
 * insertion Tx offload features.
 *
 * @param txq
 *   Pointer to TX queue structure.
 * @param loc
 *   Pointer to burst routine local context.
 * @param wqe
 *   Pointer to WQE to fill with built Ethernet Segment.
 * @param olx
 *   Configured Tx offloads mask. It is fully defined at
 *   compile time and may be used for optimization.
 */
static __rte_always_inline void
mlx5_tx_eseg_none(struct mlx5_txq_data *restrict txq __rte_unused,
		  struct mlx5_txq_local *restrict loc,
		  struct mlx5_wqe *restrict wqe,
		  unsigned int olx)
{
	struct mlx5_wqe_eseg *restrict es = &wqe->eseg;
	uint32_t csum;

	/*
	 * Calculate and set check sum flags first, dword field
	 * in segment may be shared with Software Parser flags.
	 */
	csum = MLX5_TXOFF_CONFIG(CSUM) ? txq_ol_cksum_to_cs(loc->mbuf) : 0;
	es->flags = rte_cpu_to_le_32(csum);
	/*
	 * Calculate and set Software Parser offsets and flags.
	 * These flags a set for custom UDP and IP tunnel packets.
	 */
	es->swp_offs = txq_mbuf_to_swp(loc, &es->swp_flags, olx);
	/* Fill metadata field if needed. */
	es->metadata = MLX5_TXOFF_CONFIG(METADATA) ?
		       loc->mbuf->ol_flags & PKT_TX_METADATA ?
		       loc->mbuf->tx_metadata : 0 : 0;
	/* Engage VLAN tag insertion feature if requested. */
	if (MLX5_TXOFF_CONFIG(VLAN) &&
	    loc->mbuf->ol_flags & PKT_TX_VLAN_PKT) {
		/*
		 * We should get here only if device support
		 * this feature correctly.
		 */
		assert(txq->vlan_en);
		es->inline_hdr = rte_cpu_to_be_32(MLX5_ETH_WQE_VLAN_INSERT |
						  loc->mbuf->vlan_tci);
	} else {
		es->inline_hdr = RTE_BE32(0);
	}
}

/**
 * Build the Ethernet Segment with minimal inlined data
 * of MLX5_ESEG_MIN_INLINE_SIZE bytes length. This is
 * used to fill the gap in single WQEBB WQEs.
 * Supports Software Parser, Checksums and VLAN
 * insertion Tx offload features.
 *
 * @param txq
 *   Pointer to TX queue structure.
 * @param loc
 *   Pointer to burst routine local context.
 * @param wqe
 *   Pointer to WQE to fill with built Ethernet Segment.
 * @param vlan
 *   Length of VLAN tag insertion if any.
 * @param olx
 *   Configured Tx offloads mask. It is fully defined at
 *   compile time and may be used for optimization.
 */
static __rte_always_inline void
mlx5_tx_eseg_dmin(struct mlx5_txq_data *restrict txq __rte_unused,
		  struct mlx5_txq_local *restrict loc,
		  struct mlx5_wqe *restrict wqe,
		  unsigned int vlan,
		  unsigned int olx)
{
	struct mlx5_wqe_eseg *restrict es = &wqe->eseg;
	uint32_t csum;
	uint8_t *psrc, *pdst;

	/*
	 * Calculate and set check sum flags first, dword field
	 * in segment may be shared with Software Parser flags.
	 */
	csum = MLX5_TXOFF_CONFIG(CSUM) ? txq_ol_cksum_to_cs(loc->mbuf) : 0;
	es->flags = rte_cpu_to_le_32(csum);
	/*
	 * Calculate and set Software Parser offsets and flags.
	 * These flags a set for custom UDP and IP tunnel packets.
	 */
	es->swp_offs = txq_mbuf_to_swp(loc, &es->swp_flags, olx);
	/* Fill metadata field if needed. */
	es->metadata = MLX5_TXOFF_CONFIG(METADATA) ?
		       loc->mbuf->ol_flags & PKT_TX_METADATA ?
		       loc->mbuf->tx_metadata : 0 : 0;
	static_assert(MLX5_ESEG_MIN_INLINE_SIZE ==
				(sizeof(uint16_t) +
				 sizeof(rte_v128u32_t)),
		      "invalid Ethernet Segment data size");
	static_assert(MLX5_ESEG_MIN_INLINE_SIZE ==
				(sizeof(uint16_t) +
				 sizeof(struct rte_vlan_hdr) +
				 2 * RTE_ETHER_ADDR_LEN),
		      "invalid Ethernet Segment data size");
	psrc = rte_pktmbuf_mtod(loc->mbuf, uint8_t *);
	es->inline_hdr_sz = RTE_BE16(MLX5_ESEG_MIN_INLINE_SIZE);
	es->inline_data = *(unaligned_uint16_t *)psrc;
	psrc +=	sizeof(uint16_t);
	pdst = (uint8_t *)(es + 1);
	if (MLX5_TXOFF_CONFIG(VLAN) && vlan) {
		/* Implement VLAN tag insertion as part inline data. */
		memcpy(pdst, psrc, 2 * RTE_ETHER_ADDR_LEN - sizeof(uint16_t));
		pdst += 2 * RTE_ETHER_ADDR_LEN - sizeof(uint16_t);
		psrc +=	2 * RTE_ETHER_ADDR_LEN - sizeof(uint16_t);
		/* Insert VLAN ethertype + VLAN tag. */
		*(unaligned_uint32_t *)pdst = rte_cpu_to_be_32
						((RTE_ETHER_TYPE_VLAN << 16) |
						 loc->mbuf->vlan_tci);
		pdst += sizeof(struct rte_vlan_hdr);
		/* Copy the rest two bytes from packet data. */
		assert(pdst == RTE_PTR_ALIGN(pdst, sizeof(uint16_t)));
		*(uint16_t *)pdst = *(unaligned_uint16_t *)psrc;
	} else {
		/* Fill the gap in the title WQEBB with inline data. */
		rte_mov16(pdst, psrc);
	}
}

/**
 * Build the Ethernet Segment with entire packet
 * data inlining. Checks the boundary of WQEBB and
 * ring buffer wrapping, supports Software Parser,
 * Checksums and VLAN insertion Tx offload features.
 *
 * @param txq
 *   Pointer to TX queue structure.
 * @param loc
 *   Pointer to burst routine local context.
 * @param wqe
 *   Pointer to WQE to fill with built Ethernet Segment.
 * @param vlan
 *   Length of VLAN tag insertion if any.
 * @param inlen
 *   Length of data to inline (VLAN included, if any).
 * @param tso
 *   TSO flag, set mss field from the packet.
 * @param olx
 *   Configured Tx offloads mask. It is fully defined at
 *   compile time and may be used for optimization.
 *
 * @return
 *   Pointer to the next Data Segment (aligned and wrapped around).
 */
static __rte_always_inline struct mlx5_wqe_dseg *
mlx5_tx_eseg_data(struct mlx5_txq_data *restrict txq,
		  struct mlx5_txq_local *restrict loc,
		  struct mlx5_wqe *restrict wqe,
		  unsigned int vlan,
		  unsigned int inlen,
		  unsigned int tso,
		  unsigned int olx)
{
	struct mlx5_wqe_eseg *restrict es = &wqe->eseg;
	uint32_t csum;
	uint8_t *psrc, *pdst;
	unsigned int part;

	/*
	 * Calculate and set check sum flags first, dword field
	 * in segment may be shared with Software Parser flags.
	 */
	csum = MLX5_TXOFF_CONFIG(CSUM) ? txq_ol_cksum_to_cs(loc->mbuf) : 0;
	if (tso) {
		csum <<= 24;
		csum |= loc->mbuf->tso_segsz;
		es->flags = rte_cpu_to_be_32(csum);
	} else {
		es->flags = rte_cpu_to_le_32(csum);
	}
	/*
	 * Calculate and set Software Parser offsets and flags.
	 * These flags a set for custom UDP and IP tunnel packets.
	 */
	es->swp_offs = txq_mbuf_to_swp(loc, &es->swp_flags, olx);
	/* Fill metadata field if needed. */
	es->metadata = MLX5_TXOFF_CONFIG(METADATA) ?
		       loc->mbuf->ol_flags & PKT_TX_METADATA ?
		       loc->mbuf->tx_metadata : 0 : 0;
	static_assert(MLX5_ESEG_MIN_INLINE_SIZE ==
				(sizeof(uint16_t) +
				 sizeof(rte_v128u32_t)),
		      "invalid Ethernet Segment data size");
	static_assert(MLX5_ESEG_MIN_INLINE_SIZE ==
				(sizeof(uint16_t) +
				 sizeof(struct rte_vlan_hdr) +
				 2 * RTE_ETHER_ADDR_LEN),
		      "invalid Ethernet Segment data size");
	psrc = rte_pktmbuf_mtod(loc->mbuf, uint8_t *);
	es->inline_hdr_sz = rte_cpu_to_be_16(inlen);
	es->inline_data = *(unaligned_uint16_t *)psrc;
	psrc +=	sizeof(uint16_t);
	pdst = (uint8_t *)(es + 1);
	if (MLX5_TXOFF_CONFIG(VLAN) && vlan) {
		/* Implement VLAN tag insertion as part inline data. */
		memcpy(pdst, psrc, 2 * RTE_ETHER_ADDR_LEN - sizeof(uint16_t));
		pdst += 2 * RTE_ETHER_ADDR_LEN - sizeof(uint16_t);
		psrc +=	2 * RTE_ETHER_ADDR_LEN - sizeof(uint16_t);
		/* Insert VLAN ethertype + VLAN tag. */
		*(unaligned_uint32_t *)pdst = rte_cpu_to_be_32
						((RTE_ETHER_TYPE_VLAN << 16) |
						 loc->mbuf->vlan_tci);
		pdst += sizeof(struct rte_vlan_hdr);
		/* Copy the rest two bytes from packet data. */
		assert(pdst == RTE_PTR_ALIGN(pdst, sizeof(uint16_t)));
		*(uint16_t *)pdst = *(unaligned_uint16_t *)psrc;
		psrc += sizeof(uint16_t);
	} else {
		/* Fill the gap in the title WQEBB with inline data. */
		rte_mov16(pdst, psrc);
		psrc += sizeof(rte_v128u32_t);
	}
	pdst = (uint8_t *)(es + 2);
	assert(inlen >= MLX5_ESEG_MIN_INLINE_SIZE);
	assert(pdst < (uint8_t *)txq->wqes_end);
	inlen -= MLX5_ESEG_MIN_INLINE_SIZE;
	if (!inlen) {
		assert(pdst == RTE_PTR_ALIGN(pdst, MLX5_WSEG_SIZE));
		return (struct mlx5_wqe_dseg *)pdst;
	}
	/*
	 * The WQEBB space availability is checked by caller.
	 * Here we should be aware of WQE ring buffer wraparound only.
	 */
	part = (uint8_t *)txq->wqes_end - pdst;
	part = RTE_MIN(part, inlen);
	do {
		rte_memcpy(pdst, psrc, part);
		inlen -= part;
		if (likely(!inlen)) {
			/*
			 * If return value is not used by the caller
			 * the code below will be optimized out.
			 */
			pdst += part;
			pdst = RTE_PTR_ALIGN(pdst, MLX5_WSEG_SIZE);
			if (unlikely(pdst >= (uint8_t *)txq->wqes_end))
				pdst = (uint8_t *)txq->wqes;
			return (struct mlx5_wqe_dseg *)pdst;
		}
		pdst = (uint8_t *)txq->wqes;
		psrc += part;
		part = inlen;
	} while (true);
}

/**
 * Copy data from chain of mbuf to the specified linear buffer.
 * Checksums and VLAN insertion Tx offload features. If data
 * from some mbuf copied completely this mbuf is freed. Local
 * structure is used to keep the byte stream state.
 *
 * @param pdst
 *   Pointer to the destination linear buffer.
 * @param loc
 *   Pointer to burst routine local context.
 * @param len
 *   Length of data to be copied.
 * @param olx
 *   Configured Tx offloads mask. It is fully defined at
 *   compile time and may be used for optimization.
 */
static __rte_always_inline void
mlx5_tx_mseg_memcpy(uint8_t *pdst,
		    struct mlx5_txq_local *restrict loc,
		    unsigned int len,
		    unsigned int olx __rte_unused)
{
	struct rte_mbuf *mbuf;
	unsigned int part, dlen;
	uint8_t *psrc;

	assert(len);
	do {
		/* Allow zero length packets, must check first. */
		dlen = rte_pktmbuf_data_len(loc->mbuf);
		if (dlen <= loc->mbuf_off) {
			/* Exhausted packet, just free. */
			mbuf = loc->mbuf;
			loc->mbuf = mbuf->next;
			rte_pktmbuf_free_seg(mbuf);
			loc->mbuf_off = 0;
			assert(loc->mbuf_nseg > 1);
			assert(loc->mbuf);
			--loc->mbuf_nseg;
			continue;
		}
		dlen -= loc->mbuf_off;
		psrc = rte_pktmbuf_mtod_offset(loc->mbuf, uint8_t *,
					       loc->mbuf_off);
		part = RTE_MIN(len, dlen);
		rte_memcpy(pdst, psrc, part);
		loc->mbuf_off += part;
		len -= part;
		if (!len) {
			if (loc->mbuf_off >= rte_pktmbuf_data_len(loc->mbuf)) {
				loc->mbuf_off = 0;
				/* Exhausted packet, just free. */
				mbuf = loc->mbuf;
				loc->mbuf = mbuf->next;
				rte_pktmbuf_free_seg(mbuf);
				loc->mbuf_off = 0;
				assert(loc->mbuf_nseg >= 1);
				--loc->mbuf_nseg;
			}
			return;
		}
		pdst += part;
	} while (true);
}

/**
 * Build the Ethernet Segment with inlined data from
 * multi-segment packet. Checks the boundary of WQEBB
 * and ring buffer wrapping, supports Software Parser,
 * Checksums and VLAN insertion Tx offload features.
 *
 * @param txq
 *   Pointer to TX queue structure.
 * @param loc
 *   Pointer to burst routine local context.
 * @param wqe
 *   Pointer to WQE to fill with built Ethernet Segment.
 * @param vlan
 *   Length of VLAN tag insertion if any.
 * @param inlen
 *   Length of data to inline (VLAN included, if any).
 * @param tso
 *   TSO flag, set mss field from the packet.
 * @param olx
 *   Configured Tx offloads mask. It is fully defined at
 *   compile time and may be used for optimization.
 *
 * @return
 *   Pointer to the next Data Segment (aligned and
 *   possible NOT wrapped around - caller should do
 *   wrapping check on its own).
 */
static __rte_always_inline struct mlx5_wqe_dseg *
mlx5_tx_eseg_mdat(struct mlx5_txq_data *restrict txq,
		  struct mlx5_txq_local *restrict loc,
		  struct mlx5_wqe *restrict wqe,
		  unsigned int vlan,
		  unsigned int inlen,
		  unsigned int tso,
		  unsigned int olx)
{
	struct mlx5_wqe_eseg *restrict es = &wqe->eseg;
	uint32_t csum;
	uint8_t *pdst;
	unsigned int part;

	/*
	 * Calculate and set check sum flags first, uint32_t field
	 * in segment may be shared with Software Parser flags.
	 */
	csum = MLX5_TXOFF_CONFIG(CSUM) ? txq_ol_cksum_to_cs(loc->mbuf) : 0;
	if (tso) {
		csum <<= 24;
		csum |= loc->mbuf->tso_segsz;
		es->flags = rte_cpu_to_be_32(csum);
	} else {
		es->flags = rte_cpu_to_le_32(csum);
	}
	/*
	 * Calculate and set Software Parser offsets and flags.
	 * These flags a set for custom UDP and IP tunnel packets.
	 */
	es->swp_offs = txq_mbuf_to_swp(loc, &es->swp_flags, olx);
	/* Fill metadata field if needed. */
	es->metadata = MLX5_TXOFF_CONFIG(METADATA) ?
		       loc->mbuf->ol_flags & PKT_TX_METADATA ?
		       loc->mbuf->tx_metadata : 0 : 0;
	static_assert(MLX5_ESEG_MIN_INLINE_SIZE ==
				(sizeof(uint16_t) +
				 sizeof(rte_v128u32_t)),
		      "invalid Ethernet Segment data size");
	static_assert(MLX5_ESEG_MIN_INLINE_SIZE ==
				(sizeof(uint16_t) +
				 sizeof(struct rte_vlan_hdr) +
				 2 * RTE_ETHER_ADDR_LEN),
		      "invalid Ethernet Segment data size");
	assert(inlen >= MLX5_ESEG_MIN_INLINE_SIZE);
	es->inline_hdr_sz = rte_cpu_to_be_16(inlen);
	pdst = (uint8_t *)&es->inline_data;
	if (MLX5_TXOFF_CONFIG(VLAN) && vlan) {
		/* Implement VLAN tag insertion as part inline data. */
		mlx5_tx_mseg_memcpy(pdst, loc, 2 * RTE_ETHER_ADDR_LEN, olx);
		pdst += 2 * RTE_ETHER_ADDR_LEN;
		*(unaligned_uint32_t *)pdst = rte_cpu_to_be_32
						((RTE_ETHER_TYPE_VLAN << 16) |
						 loc->mbuf->vlan_tci);
		pdst += sizeof(struct rte_vlan_hdr);
		inlen -= 2 * RTE_ETHER_ADDR_LEN + sizeof(struct rte_vlan_hdr);
	}
	assert(pdst < (uint8_t *)txq->wqes_end);
	/*
	 * The WQEBB space availability is checked by caller.
	 * Here we should be aware of WQE ring buffer wraparound only.
	 */
	part = (uint8_t *)txq->wqes_end - pdst;
	part = RTE_MIN(part, inlen);
	assert(part);
	do {
		mlx5_tx_mseg_memcpy(pdst, loc, part, olx);
		inlen -= part;
		if (likely(!inlen)) {
			pdst += part;
			pdst = RTE_PTR_ALIGN(pdst, MLX5_WSEG_SIZE);
			return (struct mlx5_wqe_dseg *)pdst;
		}
		pdst = (uint8_t *)txq->wqes;
		part = inlen;
	} while (true);
}

/**
 * Build the Data Segment of pointer type.
 *
 * @param txq
 *   Pointer to TX queue structure.
 * @param loc
 *   Pointer to burst routine local context.
 * @param dseg
 *   Pointer to WQE to fill with built Data Segment.
 * @param buf
 *   Data buffer to point.
 * @param len
 *   Data buffer length.
 * @param olx
 *   Configured Tx offloads mask. It is fully defined at
 *   compile time and may be used for optimization.
 */
static __rte_always_inline void
mlx5_tx_dseg_ptr(struct mlx5_txq_data *restrict txq,
		 struct mlx5_txq_local *restrict loc,
		 struct mlx5_wqe_dseg *restrict dseg,
		 uint8_t *buf,
		 unsigned int len,
		 unsigned int olx __rte_unused)

{
	assert(len);
	dseg->bcount = rte_cpu_to_be_32(len);
	dseg->lkey = mlx5_tx_mb2mr(txq, loc->mbuf);
	dseg->pbuf = rte_cpu_to_be_64((uintptr_t)buf);
}

/**
 * Build the Data Segment of pointer type or inline
 * if data length is less than buffer in minimal
 * Data Segment size.
 *
 * @param txq
 *   Pointer to TX queue structure.
 * @param loc
 *   Pointer to burst routine local context.
 * @param dseg
 *   Pointer to WQE to fill with built Data Segment.
 * @param buf
 *   Data buffer to point.
 * @param len
 *   Data buffer length.
 * @param olx
 *   Configured Tx offloads mask. It is fully defined at
 *   compile time and may be used for optimization.
 */
static __rte_always_inline void
mlx5_tx_dseg_iptr(struct mlx5_txq_data *restrict txq,
		  struct mlx5_txq_local *restrict loc,
		  struct mlx5_wqe_dseg *restrict dseg,
		  uint8_t *buf,
		  unsigned int len,
		  unsigned int olx __rte_unused)

{
	uintptr_t dst, src;

	assert(len);
	if (len > MLX5_DSEG_MIN_INLINE_SIZE) {
		dseg->bcount = rte_cpu_to_be_32(len);
		dseg->lkey = mlx5_tx_mb2mr(txq, loc->mbuf);
		dseg->pbuf = rte_cpu_to_be_64((uintptr_t)buf);

		return;
	}
	dseg->bcount = rte_cpu_to_be_32(len | MLX5_ETH_WQE_DATA_INLINE);
	/* Unrolled implementation of generic rte_memcpy. */
	dst = (uintptr_t)&dseg->inline_data[0];
	src = (uintptr_t)buf;
#ifdef RTE_ARCH_STRICT_ALIGN
	memcpy(dst, src, len);
#else
	if (len & 0x08) {
		*(uint64_t *)dst = *(uint64_t *)src;
		dst += sizeof(uint64_t);
		src += sizeof(uint64_t);
	}
	if (len & 0x04) {
		*(uint32_t *)dst = *(uint32_t *)src;
		dst += sizeof(uint32_t);
		src += sizeof(uint32_t);
	}
	if (len & 0x02) {
		*(uint16_t *)dst = *(uint16_t *)src;
		dst += sizeof(uint16_t);
		src += sizeof(uint16_t);
	}
	if (len & 0x01)
		*(uint8_t *)dst = *(uint8_t *)src;
#endif
}

/**
 * Build the Data Segment of inlined data from single
 * segment packet, no VLAN insertion.
 *
 * @param txq
 *   Pointer to TX queue structure.
 * @param loc
 *   Pointer to burst routine local context.
 * @param dseg
 *   Pointer to WQE to fill with built Data Segment.
 * @param buf
 *   Data buffer to point.
 * @param len
 *   Data buffer length.
 * @param olx
 *   Configured Tx offloads mask. It is fully defined at
 *   compile time and may be used for optimization.
 *
 * @return
 *   Pointer to the next Data Segment after inlined data.
 *   Ring buffer wraparound check is needed. We do not
 *   do it here because it may not be needed for the
 *   last packet in the eMPW session.
 */
static __rte_always_inline struct mlx5_wqe_dseg *
mlx5_tx_dseg_empw(struct mlx5_txq_data *restrict txq,
		  struct mlx5_txq_local *restrict loc __rte_unused,
		  struct mlx5_wqe_dseg *restrict dseg,
		  uint8_t *buf,
		  unsigned int len,
		  unsigned int olx __rte_unused)
{
	unsigned int part;
	uint8_t *pdst;

	dseg->bcount = rte_cpu_to_be_32(len | MLX5_ETH_WQE_DATA_INLINE);
	pdst = &dseg->inline_data[0];
	/*
	 * The WQEBB space availability is checked by caller.
	 * Here we should be aware of WQE ring buffer wraparound only.
	 */
	part = (uint8_t *)txq->wqes_end - pdst;
	part = RTE_MIN(part, len);
	do {
		rte_memcpy(pdst, buf, part);
		len -= part;
		if (likely(!len)) {
			pdst += part;
			pdst = RTE_PTR_ALIGN(pdst, MLX5_WSEG_SIZE);
			/* Note: no final wraparound check here. */
			return (struct mlx5_wqe_dseg *)pdst;
		}
		pdst = (uint8_t *)txq->wqes;
		buf += part;
		part = len;
	} while (true);
}

/**
 * Build the Data Segment of inlined data from single
 * segment packet with VLAN insertion.
 *
 * @param txq
 *   Pointer to TX queue structure.
 * @param loc
 *   Pointer to burst routine local context.
 * @param dseg
 *   Pointer to the dseg fill with built Data Segment.
 * @param buf
 *   Data buffer to point.
 * @param len
 *   Data buffer length.
 * @param olx
 *   Configured Tx offloads mask. It is fully defined at
 *   compile time and may be used for optimization.
 *
 * @return
 *   Pointer to the next Data Segment after inlined data.
 *   Ring buffer wraparound check is needed.
 */
static __rte_always_inline struct mlx5_wqe_dseg *
mlx5_tx_dseg_vlan(struct mlx5_txq_data *restrict txq,
		  struct mlx5_txq_local *restrict loc __rte_unused,
		  struct mlx5_wqe_dseg *restrict dseg,
		  uint8_t *buf,
		  unsigned int len,
		  unsigned int olx __rte_unused)

{
	unsigned int part;
	uint8_t *pdst;

	assert(len > MLX5_ESEG_MIN_INLINE_SIZE);
	static_assert(MLX5_DSEG_MIN_INLINE_SIZE ==
				 (2 * RTE_ETHER_ADDR_LEN),
		      "invalid Data Segment data size");
	dseg->bcount = rte_cpu_to_be_32((len + sizeof(struct rte_vlan_hdr)) |
					MLX5_ETH_WQE_DATA_INLINE);
	pdst = &dseg->inline_data[0];
	memcpy(pdst, buf, MLX5_DSEG_MIN_INLINE_SIZE);
	buf += MLX5_DSEG_MIN_INLINE_SIZE;
	pdst += MLX5_DSEG_MIN_INLINE_SIZE;
	/* Insert VLAN ethertype + VLAN tag. Pointer is aligned. */
	assert(pdst == RTE_PTR_ALIGN(pdst, MLX5_WSEG_SIZE));
	*(uint32_t *)pdst = rte_cpu_to_be_32((RTE_ETHER_TYPE_VLAN << 16) |
					      loc->mbuf->vlan_tci);
	pdst += sizeof(struct rte_vlan_hdr);
	if (unlikely(pdst >= (uint8_t *)txq->wqes_end))
		pdst = (uint8_t *)txq->wqes;
	/*
	 * The WQEBB space availability is checked by caller.
	 * Here we should be aware of WQE ring buffer wraparound only.
	 */
	part = (uint8_t *)txq->wqes_end - pdst;
	part = RTE_MIN(part, len);
	do {
		rte_memcpy(pdst, buf, part);
		len -= part;
		if (likely(!len)) {
			pdst += part;
			pdst = RTE_PTR_ALIGN(pdst, MLX5_WSEG_SIZE);
			/* Note: no final wraparound check here. */
			return (struct mlx5_wqe_dseg *)pdst;
		}
		pdst = (uint8_t *)txq->wqes;
		buf += part;
		part = len;
	} while (true);
}

/**
 * Build the Ethernet Segment with optionally inlined data with
 * VLAN insertion and following Data Segments (if any) from
 * multi-segment packet. Used by ordinary send and TSO.
 *
 * @param txq
 *   Pointer to TX queue structure.
 * @param loc
 *   Pointer to burst routine local context.
 * @param wqe
 *   Pointer to WQE to fill with built Ethernet/Data Segments.
 * @param vlan
 *   Length of VLAN header to insert, 0 means no VLAN insertion.
 * @param inlen
 *   Data length to inline. For TSO this parameter specifies
 *   exact value, for ordinary send routine can be aligned by
 *   caller to provide better WQE space saving and data buffer
 *   start address alignment. This length includes VLAN header
 *   being inserted.
 * @param tso
 *   Zero means ordinary send, inlined data can be extended,
 *   otherwise this is TSO, inlined data length is fixed.
 * @param olx
 *   Configured Tx offloads mask. It is fully defined at
 *   compile time and may be used for optimization.
 *
 * @return
 *   Actual size of built WQE in segments.
 */
static __rte_always_inline unsigned int
mlx5_tx_mseg_build(struct mlx5_txq_data *restrict txq,
		   struct mlx5_txq_local *restrict loc,
		   struct mlx5_wqe *restrict wqe,
		   unsigned int vlan,
		   unsigned int inlen,
		   unsigned int tso,
		   unsigned int olx __rte_unused)
{
	struct mlx5_wqe_dseg *restrict dseg;
	unsigned int ds;

	assert((rte_pktmbuf_pkt_len(loc->mbuf) + vlan) >= inlen);
	loc->mbuf_nseg = NB_SEGS(loc->mbuf);
	loc->mbuf_off = 0;

	dseg = mlx5_tx_eseg_mdat(txq, loc, wqe, vlan, inlen, tso, olx);
	if (!loc->mbuf_nseg)
		goto dseg_done;
	/*
	 * There are still some mbuf remaining, not inlined.
	 * The first mbuf may be partially inlined and we
	 * must process the possible non-zero data offset.
	 */
	if (loc->mbuf_off) {
		unsigned int dlen;
		uint8_t *dptr;

		/*
		 * Exhausted packets must be dropped before.
		 * Non-zero offset means there are some data
		 * remained in the packet.
		 */
		assert(loc->mbuf_off < rte_pktmbuf_data_len(loc->mbuf));
		assert(rte_pktmbuf_data_len(loc->mbuf));
		dptr = rte_pktmbuf_mtod_offset(loc->mbuf, uint8_t *,
					       loc->mbuf_off);
		dlen = rte_pktmbuf_data_len(loc->mbuf) - loc->mbuf_off;
		/*
		 * Build the pointer/minimal data Data Segment.
		 * Do ring buffer wrapping check in advance.
		 */
		if ((uintptr_t)dseg >= (uintptr_t)txq->wqes_end)
			dseg = (struct mlx5_wqe_dseg *)txq->wqes;
		mlx5_tx_dseg_iptr(txq, loc, dseg, dptr, dlen, olx);
		/* Store the mbuf to be freed on completion. */
		assert(loc->elts_free);
		txq->elts[txq->elts_head++ & txq->elts_m] = loc->mbuf;
		--loc->elts_free;
		++dseg;
		if (--loc->mbuf_nseg == 0)
			goto dseg_done;
		loc->mbuf = loc->mbuf->next;
		loc->mbuf_off = 0;
	}
	do {
		if (unlikely(!rte_pktmbuf_data_len(loc->mbuf))) {
			struct rte_mbuf *mbuf;

			/* Zero length segment found, just skip. */
			mbuf = loc->mbuf;
			loc->mbuf = loc->mbuf->next;
			rte_pktmbuf_free_seg(mbuf);
			if (--loc->mbuf_nseg == 0)
				break;
		} else {
			if ((uintptr_t)dseg >= (uintptr_t)txq->wqes_end)
				dseg = (struct mlx5_wqe_dseg *)txq->wqes;
			mlx5_tx_dseg_iptr
				(txq, loc, dseg,
				 rte_pktmbuf_mtod(loc->mbuf, uint8_t *),
				 rte_pktmbuf_data_len(loc->mbuf), olx);
			assert(loc->elts_free);
			txq->elts[txq->elts_head++ & txq->elts_m] = loc->mbuf;
			--loc->elts_free;
			++dseg;
			if (--loc->mbuf_nseg == 0)
				break;
			loc->mbuf = loc->mbuf->next;
		}
	} while (true);

dseg_done:
	/* Calculate actual segments used from the dseg pointer. */
	if ((uintptr_t)wqe < (uintptr_t)dseg)
		ds = ((uintptr_t)dseg - (uintptr_t)wqe) / MLX5_WSEG_SIZE;
	else
		ds = (((uintptr_t)dseg - (uintptr_t)wqe) +
		      txq->wqe_s * MLX5_WQE_SIZE) / MLX5_WSEG_SIZE;
	return ds;
}

/**
 * Tx one packet function for multi-segment TSO. Supports all
 * types of Tx offloads, uses MLX5_OPCODE_TSO to build WQEs,
 * sends one packet per WQE.
 *
 * This routine is responsible for storing processed mbuf
 * into elts ring buffer and update elts_head.
 *
 * @param txq
 *   Pointer to TX queue structure.
 * @param loc
 *   Pointer to burst routine local context.
 * @param olx
 *   Configured Tx offloads mask. It is fully defined at
 *   compile time and may be used for optimization.
 *
 * @return
 *   MLX5_TXCMP_CODE_EXIT - sending is done or impossible.
 *   MLX5_TXCMP_CODE_ERROR - some unrecoverable error occurred.
 * Local context variables partially updated.
 */
static __rte_always_inline enum mlx5_txcmp_code
mlx5_tx_packet_multi_tso(struct mlx5_txq_data *restrict txq,
			struct mlx5_txq_local *restrict loc,
			unsigned int olx)
{
	struct mlx5_wqe *restrict wqe;
	unsigned int ds, dlen, inlen, ntcp, vlan = 0;

	/*
	 * Calculate data length to be inlined to estimate
	 * the required space in WQE ring buffer.
	 */
	dlen = rte_pktmbuf_pkt_len(loc->mbuf);
	if (MLX5_TXOFF_CONFIG(VLAN) && loc->mbuf->ol_flags & PKT_TX_VLAN_PKT)
		vlan = sizeof(struct rte_vlan_hdr);
	inlen = loc->mbuf->l2_len + vlan +
		loc->mbuf->l3_len + loc->mbuf->l4_len;
	if (unlikely((!inlen || !loc->mbuf->tso_segsz)))
		return MLX5_TXCMP_CODE_ERROR;
	if (loc->mbuf->ol_flags & PKT_TX_TUNNEL_MASK)
		inlen += loc->mbuf->outer_l2_len + loc->mbuf->outer_l3_len;
	/* Packet must contain all TSO headers. */
	if (unlikely(inlen > MLX5_MAX_TSO_HEADER ||
		     inlen <= MLX5_ESEG_MIN_INLINE_SIZE ||
		     inlen > (dlen + vlan)))
		return MLX5_TXCMP_CODE_ERROR;
	assert(inlen >= txq->inlen_mode);
	/*
	 * Check whether there are enough free WQEBBs:
	 * - Control Segment
	 * - Ethernet Segment
	 * - First Segment of inlined Ethernet data
	 * - ... data continued ...
	 * - Data Segments of pointer/min inline type
	 */
	ds = NB_SEGS(loc->mbuf) + 2 + (inlen -
				       MLX5_ESEG_MIN_INLINE_SIZE +
				       MLX5_WSEG_SIZE +
				       MLX5_WSEG_SIZE - 1) / MLX5_WSEG_SIZE;
	if (unlikely(loc->wqe_free < ((ds + 3) / 4)))
		return MLX5_TXCMP_CODE_EXIT;
	/* Check for maximal WQE size. */
	if (unlikely((MLX5_WQE_SIZE_MAX / MLX5_WSEG_SIZE) < ((ds + 3) / 4)))
		return MLX5_TXCMP_CODE_ERROR;
#ifdef MLX5_PMD_SOFT_COUNTERS
	/* Update sent data bytes/packets counters. */
	ntcp = (dlen - (inlen - vlan) + loc->mbuf->tso_segsz - 1) /
		loc->mbuf->tso_segsz;
	/*
	 * One will be added for mbuf itself
	 * at the end of the mlx5_tx_burst from
	 * loc->pkts_sent field.
	 */
	--ntcp;
	txq->stats.opackets += ntcp;
	txq->stats.obytes += dlen + vlan + ntcp * inlen;
#endif
	wqe = txq->wqes + (txq->wqe_ci & txq->wqe_m);
	loc->wqe_last = wqe;
	mlx5_tx_cseg_init(txq, loc, wqe, 0, MLX5_OPCODE_TSO, olx);
	ds = mlx5_tx_mseg_build(txq, loc, wqe, vlan, inlen, 1, olx);
	wqe->cseg.sq_ds = rte_cpu_to_be_32(txq->qp_num_8s | ds);
	txq->wqe_ci += (ds + 3) / 4;
	loc->wqe_free -= (ds + 3) / 4;
	/* Request CQE generation if limits are reached. */
	mlx5_tx_request_completion(txq, loc, true, olx);
	return MLX5_TXCMP_CODE_MULTI;
}

/**
 * Tx one packet function for multi-segment SEND. Supports all
 * types of Tx offloads, uses MLX5_OPCODE_SEND to build WQEs,
 * sends one packet per WQE, without any data inlining in
 * Ethernet Segment.
 *
 * This routine is responsible for storing processed mbuf
 * into elts ring buffer and update elts_head.
 *
 * @param txq
 *   Pointer to TX queue structure.
 * @param loc
 *   Pointer to burst routine local context.
 * @param olx
 *   Configured Tx offloads mask. It is fully defined at
 *   compile time and may be used for optimization.
 *
 * @return
 *   MLX5_TXCMP_CODE_EXIT - sending is done or impossible.
 *   MLX5_TXCMP_CODE_ERROR - some unrecoverable error occurred.
 * Local context variables partially updated.
 */
static __rte_always_inline enum mlx5_txcmp_code
mlx5_tx_packet_multi_send(struct mlx5_txq_data *restrict txq,
			  struct mlx5_txq_local *restrict loc,
			  unsigned int olx)
{
	struct mlx5_wqe_dseg *restrict dseg;
	struct mlx5_wqe *restrict wqe;
	unsigned int ds, nseg;

	assert(NB_SEGS(loc->mbuf) > 1);
	/*
	 * No inline at all, it means the CPU cycles saving
	 * is prioritized at configuration, we should not
	 * copy any packet data to WQE.
	 */
	nseg = NB_SEGS(loc->mbuf);
	ds = 2 + nseg;
	if (unlikely(loc->wqe_free < ((ds + 3) / 4)))
		return MLX5_TXCMP_CODE_EXIT;
	/* Check for maximal WQE size. */
	if (unlikely((MLX5_WQE_SIZE_MAX / MLX5_WSEG_SIZE) < ((ds + 3) / 4)))
		return MLX5_TXCMP_CODE_ERROR;
	/*
	 * Some Tx offloads may cause an error if
	 * packet is not long enough, check against
	 * assumed minimal length.
	 */
	if (rte_pktmbuf_pkt_len(loc->mbuf) <= MLX5_ESEG_MIN_INLINE_SIZE)
		return MLX5_TXCMP_CODE_ERROR;
#ifdef MLX5_PMD_SOFT_COUNTERS
	/* Update sent data bytes counter. */
	txq->stats.obytes += rte_pktmbuf_pkt_len(loc->mbuf);
	if (MLX5_TXOFF_CONFIG(VLAN) &&
	    loc->mbuf->ol_flags & PKT_TX_VLAN_PKT)
		txq->stats.obytes += sizeof(struct rte_vlan_hdr);
#endif
	/*
	 * SEND WQE, one WQEBB:
	 * - Control Segment, SEND opcode
	 * - Ethernet Segment, optional VLAN, no inline
	 * - Data Segments, pointer only type
	 */
	wqe = txq->wqes + (txq->wqe_ci & txq->wqe_m);
	loc->wqe_last = wqe;
	mlx5_tx_cseg_init(txq, loc, wqe, ds, MLX5_OPCODE_SEND, olx);
	mlx5_tx_eseg_none(txq, loc, wqe, olx);
	dseg = &wqe->dseg[0];
	do {
		if (unlikely(!rte_pktmbuf_data_len(loc->mbuf))) {
			struct rte_mbuf *mbuf;

			/*
			 * Zero length segment found, have to
			 * correct total size of WQE in segments.
			 * It is supposed to be rare occasion, so
			 * in normal case (no zero length segments)
			 * we avoid extra writing to the Control
			 * Segment.
			 */
			--ds;
			wqe->cseg.sq_ds -= RTE_BE32(1);
			mbuf = loc->mbuf;
			loc->mbuf = mbuf->next;
			rte_pktmbuf_free_seg(mbuf);
			if (--nseg == 0)
				break;
		} else {
			mlx5_tx_dseg_ptr
				(txq, loc, dseg,
				 rte_pktmbuf_mtod(loc->mbuf, uint8_t *),
				 rte_pktmbuf_data_len(loc->mbuf), olx);
			txq->elts[txq->elts_head++ & txq->elts_m] = loc->mbuf;
			--loc->elts_free;
			if (--nseg == 0)
				break;
			++dseg;
			if ((uintptr_t)dseg >= (uintptr_t)txq->wqes_end)
				dseg = (struct mlx5_wqe_dseg *)txq->wqes;
			loc->mbuf = loc->mbuf->next;
		}
	} while (true);
	txq->wqe_ci += (ds + 3) / 4;
	loc->wqe_free -= (ds + 3) / 4;
	/* Request CQE generation if limits are reached. */
	mlx5_tx_request_completion(txq, loc, true, olx);
	return MLX5_TXCMP_CODE_MULTI;
}

/**
 * Tx one packet function for multi-segment SEND. Supports all
 * types of Tx offloads, uses MLX5_OPCODE_SEND to build WQEs,
 * sends one packet per WQE, with data inlining in
 * Ethernet Segment and minimal Data Segments.
 *
 * This routine is responsible for storing processed mbuf
 * into elts ring buffer and update elts_head.
 *
 * @param txq
 *   Pointer to TX queue structure.
 * @param loc
 *   Pointer to burst routine local context.
 * @param olx
 *   Configured Tx offloads mask. It is fully defined at
 *   compile time and may be used for optimization.
 *
 * @return
 *   MLX5_TXCMP_CODE_EXIT - sending is done or impossible.
 *   MLX5_TXCMP_CODE_ERROR - some unrecoverable error occurred.
 * Local context variables partially updated.
 */
static __rte_always_inline enum mlx5_txcmp_code
mlx5_tx_packet_multi_inline(struct mlx5_txq_data *restrict txq,
			    struct mlx5_txq_local *restrict loc,
			    unsigned int olx)
{
	struct mlx5_wqe *restrict wqe;
	unsigned int ds, inlen, dlen, vlan = 0;

	assert(MLX5_TXOFF_CONFIG(INLINE));
	assert(NB_SEGS(loc->mbuf) > 1);
	/*
	 * First calculate data length to be inlined
	 * to estimate the required space for WQE.
	 */
	dlen = rte_pktmbuf_pkt_len(loc->mbuf);
	if (MLX5_TXOFF_CONFIG(VLAN) && loc->mbuf->ol_flags & PKT_TX_VLAN_PKT)
		vlan = sizeof(struct rte_vlan_hdr);
	inlen = dlen + vlan;
	/* Check against minimal length. */
	if (inlen <= MLX5_ESEG_MIN_INLINE_SIZE)
		return MLX5_TXCMP_CODE_ERROR;
	assert(txq->inlen_send >= MLX5_ESEG_MIN_INLINE_SIZE);
	if (inlen > txq->inlen_send) {
		struct rte_mbuf *mbuf;
		unsigned int nxlen;
		uintptr_t start;

		/*
		 * Packet length exceeds the allowed inline
		 * data length, check whether the minimal
		 * inlining is required.
		 */
		if (txq->inlen_mode) {
			assert(txq->inlen_mode >= MLX5_ESEG_MIN_INLINE_SIZE);
			assert(txq->inlen_mode <= txq->inlen_send);
			inlen = txq->inlen_mode;
		} else {
			if (!vlan || txq->vlan_en) {
				/*
				 * VLAN insertion will be done inside by HW.
				 * It is not utmost effective - VLAN flag is
				 * checked twice, but we should proceed the
				 * inlining length correctly and take into
				 * account the VLAN header being inserted.
				 */
				return mlx5_tx_packet_multi_send
							(txq, loc, olx);
			}
			inlen = MLX5_ESEG_MIN_INLINE_SIZE;
		}
		/*
		 * Now we know the minimal amount of data is requested
		 * to inline. Check whether we should inline the buffers
		 * from the chain beginning to eliminate some mbufs.
		 */
		mbuf = loc->mbuf;
		nxlen = rte_pktmbuf_data_len(mbuf);
		if (unlikely(nxlen <= txq->inlen_send)) {
			/* We can inline first mbuf at least. */
			if (nxlen < inlen) {
				unsigned int smlen;

				/* Scan mbufs till inlen filled. */
				do {
					smlen = nxlen;
					mbuf = NEXT(mbuf);
					assert(mbuf);
					nxlen = rte_pktmbuf_data_len(mbuf);
					nxlen += smlen;
				} while (unlikely(nxlen < inlen));
				if (unlikely(nxlen > txq->inlen_send)) {
					/* We cannot inline entire mbuf. */
					smlen = inlen - smlen;
					start = rte_pktmbuf_mtod_offset
						    (mbuf, uintptr_t, smlen);
					goto do_align;
				}
			}
			do {
				inlen = nxlen;
				mbuf = NEXT(mbuf);
				/* There should be not end of packet. */
				assert(mbuf);
				nxlen = inlen + rte_pktmbuf_data_len(mbuf);
			} while (unlikely(nxlen < txq->inlen_send));
		}
		start = rte_pktmbuf_mtod(mbuf, uintptr_t);
		/*
		 * Check whether we can do inline to align start
		 * address of data buffer to cacheline.
		 */
do_align:
		start = (~start + 1) & (RTE_CACHE_LINE_SIZE - 1);
		if (unlikely(start)) {
			start += inlen;
			if (start <= txq->inlen_send)
				inlen = start;
		}
	}
	/*
	 * Check whether there are enough free WQEBBs:
	 * - Control Segment
	 * - Ethernet Segment
	 * - First Segment of inlined Ethernet data
	 * - ... data continued ...
	 * - Data Segments of pointer/min inline type
	 *
	 * Estimate the number of Data Segments conservatively,
	 * supposing no any mbufs is being freed during inlining.
	 */
	assert(inlen <= txq->inlen_send);
	ds = NB_SEGS(loc->mbuf) + 2 + (inlen -
				       MLX5_ESEG_MIN_INLINE_SIZE +
				       MLX5_WSEG_SIZE +
				       MLX5_WSEG_SIZE - 1) / MLX5_WSEG_SIZE;
	if (unlikely(loc->wqe_free < ((ds + 3) / 4)))
		return MLX5_TXCMP_CODE_EXIT;
	/* Check for maximal WQE size. */
	if (unlikely((MLX5_WQE_SIZE_MAX / MLX5_WSEG_SIZE) < ((ds + 3) / 4)))
		return MLX5_TXCMP_CODE_ERROR;
#ifdef MLX5_PMD_SOFT_COUNTERS
	/* Update sent data bytes/packets counters. */
	txq->stats.obytes += dlen + vlan;
#endif
	wqe = txq->wqes + (txq->wqe_ci & txq->wqe_m);
	loc->wqe_last = wqe;
	mlx5_tx_cseg_init(txq, loc, wqe, 0, MLX5_OPCODE_SEND, olx);
	ds = mlx5_tx_mseg_build(txq, loc, wqe, vlan, inlen, 0, olx);
	wqe->cseg.sq_ds = rte_cpu_to_be_32(txq->qp_num_8s | ds);
	txq->wqe_ci += (ds + 3) / 4;
	loc->wqe_free -= (ds + 3) / 4;
	/* Request CQE generation if limits are reached. */
	mlx5_tx_request_completion(txq, loc, true, olx);
	return MLX5_TXCMP_CODE_MULTI;
}

/**
 * Tx burst function for multi-segment packets. Supports all
 * types of Tx offloads, uses MLX5_OPCODE_SEND/TSO to build WQEs,
 * sends one packet per WQE. Function stops sending if it
 * encounters the single-segment packet.
 *
 * This routine is responsible for storing processed mbuf
 * into elts ring buffer and update elts_head.
 *
 * @param txq
 *   Pointer to TX queue structure.
 * @param[in] pkts
 *   Packets to transmit.
 * @param pkts_n
 *   Number of packets in array.
 * @param loc
 *   Pointer to burst routine local context.
 * @param olx
 *   Configured Tx offloads mask. It is fully defined at
 *   compile time and may be used for optimization.
 *
 * @return
 *   MLX5_TXCMP_CODE_EXIT - sending is done or impossible.
 *   MLX5_TXCMP_CODE_ERROR - some unrecoverable error occurred.
 *   MLX5_TXCMP_CODE_SINGLE - single-segment packet encountered.
 *   MLX5_TXCMP_CODE_TSO - TSO single-segment packet encountered.
 * Local context variables updated.
 */
static __rte_always_inline enum mlx5_txcmp_code
mlx5_tx_burst_mseg(struct mlx5_txq_data *restrict txq,
		   struct rte_mbuf **restrict pkts,
		   unsigned int pkts_n,
		   struct mlx5_txq_local *restrict loc,
		   unsigned int olx)
{
	assert(loc->elts_free && loc->wqe_free);
	assert(pkts_n > loc->pkts_sent);
	pkts += loc->pkts_sent + 1;
	pkts_n -= loc->pkts_sent;
	for (;;) {
		enum mlx5_txcmp_code ret;

		assert(NB_SEGS(loc->mbuf) > 1);
		/*
		 * Estimate the number of free elts quickly but
		 * conservatively. Some segment may be fully inlined
		 * and freed, ignore this here - precise estimation
		 * is costly.
		 */
		if (loc->elts_free < NB_SEGS(loc->mbuf))
			return MLX5_TXCMP_CODE_EXIT;
		if (MLX5_TXOFF_CONFIG(TSO) &&
		    unlikely(loc->mbuf->ol_flags & PKT_TX_TCP_SEG)) {
			/* Proceed with multi-segment TSO. */
			ret = mlx5_tx_packet_multi_tso(txq, loc, olx);
		} else if (MLX5_TXOFF_CONFIG(INLINE)) {
			/* Proceed with multi-segment SEND with inlining. */
			ret = mlx5_tx_packet_multi_inline(txq, loc, olx);
		} else {
			/* Proceed with multi-segment SEND w/o inlining. */
			ret = mlx5_tx_packet_multi_send(txq, loc, olx);
		}
		if (ret == MLX5_TXCMP_CODE_EXIT)
			return MLX5_TXCMP_CODE_EXIT;
		if (ret == MLX5_TXCMP_CODE_ERROR)
			return MLX5_TXCMP_CODE_ERROR;
		/* WQE is built, go to the next packet. */
		++loc->pkts_sent;
		--pkts_n;
		if (unlikely(!pkts_n || !loc->elts_free || !loc->wqe_free))
			return MLX5_TXCMP_CODE_EXIT;
		loc->mbuf = *pkts++;
		if (pkts_n > 1)
			rte_prefetch0(*pkts);
		if (likely(NB_SEGS(loc->mbuf) > 1))
			continue;
		/* Here ends the series of multi-segment packets. */
		if (MLX5_TXOFF_CONFIG(TSO) &&
		    unlikely(!(loc->mbuf->ol_flags & PKT_TX_TCP_SEG)))
			return MLX5_TXCMP_CODE_TSO;
		return MLX5_TXCMP_CODE_SINGLE;
	}
	assert(false);
}

/**
 * Tx burst function for single-segment packets with TSO.
 * Supports all types of Tx offloads, except multi-packets.
 * Uses MLX5_OPCODE_TSO to build WQEs, sends one packet per WQE.
 * Function stops sending if it encounters the multi-segment
 * packet or packet without TSO requested.
 *
 * The routine is responsible for storing processed mbuf
 * into elts ring buffer and update elts_head if inline
 * offloads is requested due to possible early freeing
 * of the inlined mbufs (can not store pkts array in elts
 * as a batch).
 *
 * @param txq
 *   Pointer to TX queue structure.
 * @param[in] pkts
 *   Packets to transmit.
 * @param pkts_n
 *   Number of packets in array.
 * @param loc
 *   Pointer to burst routine local context.
 * @param olx
 *   Configured Tx offloads mask. It is fully defined at
 *   compile time and may be used for optimization.
 *
 * @return
 *   MLX5_TXCMP_CODE_EXIT - sending is done or impossible.
 *   MLX5_TXCMP_CODE_ERROR - some unrecoverable error occurred.
 *   MLX5_TXCMP_CODE_SINGLE - single-segment packet encountered.
 *   MLX5_TXCMP_CODE_MULTI - multi-segment packet encountered.
 * Local context variables updated.
 */
static __rte_always_inline enum mlx5_txcmp_code
mlx5_tx_burst_tso(struct mlx5_txq_data *restrict txq,
		  struct rte_mbuf **restrict pkts,
		  unsigned int pkts_n,
		  struct mlx5_txq_local *restrict loc,
		  unsigned int olx)
{
	assert(loc->elts_free && loc->wqe_free);
	assert(pkts_n > loc->pkts_sent);
	pkts += loc->pkts_sent + 1;
	pkts_n -= loc->pkts_sent;
	for (;;) {
		struct mlx5_wqe_dseg *restrict dseg;
		struct mlx5_wqe *restrict wqe;
		unsigned int ds, dlen, hlen, ntcp, vlan = 0;
		uint8_t *dptr;

		assert(NB_SEGS(loc->mbuf) == 1);
		dlen = rte_pktmbuf_data_len(loc->mbuf);
		if (MLX5_TXOFF_CONFIG(VLAN) &&
		    loc->mbuf->ol_flags & PKT_TX_VLAN_PKT) {
			vlan = sizeof(struct rte_vlan_hdr);
		}
		/*
		 * First calculate the WQE size to check
		 * whether we have enough space in ring buffer.
		 */
		hlen = loc->mbuf->l2_len + vlan +
		       loc->mbuf->l3_len + loc->mbuf->l4_len;
		if (unlikely((!hlen || !loc->mbuf->tso_segsz)))
			return MLX5_TXCMP_CODE_ERROR;
		if (loc->mbuf->ol_flags & PKT_TX_TUNNEL_MASK)
			hlen += loc->mbuf->outer_l2_len +
				loc->mbuf->outer_l3_len;
		/* Segment must contain all TSO headers. */
		if (unlikely(hlen > MLX5_MAX_TSO_HEADER ||
			     hlen <= MLX5_ESEG_MIN_INLINE_SIZE ||
			     hlen > (dlen + vlan)))
			return MLX5_TXCMP_CODE_ERROR;
		/*
		 * Check whether there are enough free WQEBBs:
		 * - Control Segment
		 * - Ethernet Segment
		 * - First Segment of inlined Ethernet data
		 * - ... data continued ...
		 * - Finishing Data Segment of pointer type
		 */
		ds = 4 + (hlen - MLX5_ESEG_MIN_INLINE_SIZE +
			  MLX5_WSEG_SIZE - 1) / MLX5_WSEG_SIZE;
		if (loc->wqe_free < ((ds + 3) / 4))
			return MLX5_TXCMP_CODE_EXIT;
#ifdef MLX5_PMD_SOFT_COUNTERS
		/* Update sent data bytes/packets counters. */
		ntcp = (dlen + vlan - hlen +
			loc->mbuf->tso_segsz - 1) /
			loc->mbuf->tso_segsz;
		/*
		 * One will be added for mbuf itself at the end
		 * of the mlx5_tx_burst from loc->pkts_sent field.
		 */
		--ntcp;
		txq->stats.opackets += ntcp;
		txq->stats.obytes += dlen + vlan + ntcp * hlen;
#endif
		/*
		 * Build the TSO WQE:
		 * - Control Segment
		 * - Ethernet Segment with hlen bytes inlined
		 * - Data Segment of pointer type
		 */
		wqe = txq->wqes + (txq->wqe_ci & txq->wqe_m);
		loc->wqe_last = wqe;
		mlx5_tx_cseg_init(txq, loc, wqe, ds,
				  MLX5_OPCODE_TSO, olx);
		dseg = mlx5_tx_eseg_data(txq, loc, wqe, vlan, hlen, 1, olx);
		dptr = rte_pktmbuf_mtod(loc->mbuf, uint8_t *) + hlen - vlan;
		dlen -= hlen - vlan;
		mlx5_tx_dseg_ptr(txq, loc, dseg, dptr, dlen, olx);
		/*
		 * WQE is built, update the loop parameters
		 * and go to the next packet.
		 */
		txq->wqe_ci += (ds + 3) / 4;
		loc->wqe_free -= (ds + 3) / 4;
		if (MLX5_TXOFF_CONFIG(INLINE))
			txq->elts[txq->elts_head++ & txq->elts_m] = loc->mbuf;
		--loc->elts_free;
		++loc->pkts_sent;
		--pkts_n;
		/* Request CQE generation if limits are reached. */
		mlx5_tx_request_completion(txq, loc, false, olx);
		if (unlikely(!pkts_n || !loc->elts_free || !loc->wqe_free))
			return MLX5_TXCMP_CODE_EXIT;
		loc->mbuf = *pkts++;
		if (pkts_n > 1)
			rte_prefetch0(*pkts);
		if (MLX5_TXOFF_CONFIG(MULTI) &&
		    unlikely(NB_SEGS(loc->mbuf) > 1))
			return MLX5_TXCMP_CODE_MULTI;
		if (unlikely(!(loc->mbuf->ol_flags & PKT_TX_TCP_SEG)))
			return MLX5_TXCMP_CODE_SINGLE;
		/* Continue with the next TSO packet. */
	}
	assert(false);
}

/**
 * Analyze the packet and select the best method to send.
 *
 * @param txq
 *   Pointer to TX queue structure.
 * @param loc
 *   Pointer to burst routine local context.
 * @param olx
 *   Configured Tx offloads mask. It is fully defined at
 *   compile time and may be used for optimization.
 * @param newp
 *   The predefined flag whether do complete check for
 *   multi-segment packets and TSO.
 *
 * @return
 *  MLX5_TXCMP_CODE_MULTI - multi-segment packet encountered.
 *  MLX5_TXCMP_CODE_TSO - TSO required, use TSO/LSO.
 *  MLX5_TXCMP_CODE_SINGLE - single-segment packet, use SEND.
 *  MLX5_TXCMP_CODE_EMPW - single-segment packet, use MPW.
 */
static __rte_always_inline enum mlx5_txcmp_code
mlx5_tx_able_to_empw(struct mlx5_txq_data *restrict txq,
		     struct mlx5_txq_local *restrict loc,
		     unsigned int olx,
		     bool newp)
{
	/* Check for multi-segment packet. */
	if (newp &&
	    MLX5_TXOFF_CONFIG(MULTI) &&
	    unlikely(NB_SEGS(loc->mbuf) > 1))
		return MLX5_TXCMP_CODE_MULTI;
	/* Check for TSO packet. */
	if (newp &&
	    MLX5_TXOFF_CONFIG(TSO) &&
	    unlikely(loc->mbuf->ol_flags & PKT_TX_TCP_SEG))
		return MLX5_TXCMP_CODE_TSO;
	/* Check if eMPW is enabled at all. */
	if (!MLX5_TXOFF_CONFIG(EMPW))
		return MLX5_TXCMP_CODE_SINGLE;
	/* Check if eMPW can be engaged. */
	if (MLX5_TXOFF_CONFIG(VLAN) &&
	    unlikely(loc->mbuf->ol_flags & PKT_TX_VLAN_PKT) &&
		(!MLX5_TXOFF_CONFIG(INLINE) ||
		 unlikely((rte_pktmbuf_data_len(loc->mbuf) +
			   sizeof(struct rte_vlan_hdr)) > txq->inlen_empw))) {
		/*
		 * eMPW does not support VLAN insertion offload,
		 * we have to inline the entire packet but
		 * packet is too long for inlining.
		 */
		return MLX5_TXCMP_CODE_SINGLE;
	}
	return MLX5_TXCMP_CODE_EMPW;
}

/**
 * Check the next packet attributes to match with the eMPW batch ones.
 *
 * @param txq
 *   Pointer to TX queue structure.
 * @param es
 *   Pointer to Ethernet Segment of eMPW batch.
 * @param loc
 *   Pointer to burst routine local context.
 * @param olx
 *   Configured Tx offloads mask. It is fully defined at
 *   compile time and may be used for optimization.
 *
 * @return
 *  true - packet match with eMPW batch attributes.
 *  false - no match, eMPW should be restarted.
 */
static __rte_always_inline bool
mlx5_tx_match_empw(struct mlx5_txq_data *restrict txq __rte_unused,
		   struct mlx5_wqe_eseg *restrict es,
		   struct mlx5_txq_local *restrict loc,
		   unsigned int olx)
{
	uint8_t swp_flags = 0;

	/* Compare the checksum flags, if any. */
	if (MLX5_TXOFF_CONFIG(CSUM) &&
	    txq_ol_cksum_to_cs(loc->mbuf) != es->cs_flags)
		return false;
	/* Compare the Software Parser offsets and flags. */
	if (MLX5_TXOFF_CONFIG(SWP) &&
	    (es->swp_offs != txq_mbuf_to_swp(loc, &swp_flags, olx) ||
	     es->swp_flags != swp_flags))
		return false;
	/* Fill metadata field if needed. */
	if (MLX5_TXOFF_CONFIG(METADATA) &&
		es->metadata != (loc->mbuf->ol_flags & PKT_TX_METADATA ?
				 loc->mbuf->tx_metadata : 0))
		return false;
	/* There must be no VLAN packets in eMPW loop. */
	if (MLX5_TXOFF_CONFIG(VLAN))
		assert(!(loc->mbuf->ol_flags & PKT_TX_VLAN_PKT));
	return true;
}

/*
 * Update send loop variables and WQE for eMPW loop
 * without data inlining. Number of Data Segments is
 * equal to the number of sent packets.
 *
 * @param txq
 *   Pointer to TX queue structure.
 * @param loc
 *   Pointer to burst routine local context.
 * @param ds
 *   Number of packets/Data Segments/Packets.
 * @param slen
 *   Accumulated statistics, bytes sent
 * @param olx
 *   Configured Tx offloads mask. It is fully defined at
 *   compile time and may be used for optimization.
 *
 * @return
 *  true - packet match with eMPW batch attributes.
 *  false - no match, eMPW should be restarted.
 */
static __rte_always_inline void
mlx5_tx_sdone_empw(struct mlx5_txq_data *restrict txq,
		   struct mlx5_txq_local *restrict loc,
		   unsigned int ds,
		   unsigned int slen,
		   unsigned int olx)
{
	assert(!MLX5_TXOFF_CONFIG(INLINE));
#ifdef MLX5_PMD_SOFT_COUNTERS
	/* Update sent data bytes counter. */
	 txq->stats.obytes += slen;
#else
	(void)slen;
#endif
	loc->elts_free -= ds;
	loc->pkts_sent += ds;
	ds += 2;
	loc->wqe_last->cseg.sq_ds = rte_cpu_to_be_32(txq->qp_num_8s | ds);
	txq->wqe_ci += (ds + 3) / 4;
	loc->wqe_free -= (ds + 3) / 4;
	/* Request CQE generation if limits are reached. */
	mlx5_tx_request_completion(txq, loc, false, olx);
}

/*
 * Update send loop variables and WQE for eMPW loop
 * with data inlining. Gets the size of pushed descriptors
 * and data to the WQE.
 *
 * @param txq
 *   Pointer to TX queue structure.
 * @param loc
 *   Pointer to burst routine local context.
 * @param len
 *   Total size of descriptor/data in bytes.
 * @param slen
 *   Accumulated statistics, data bytes sent.
 * @param olx
 *   Configured Tx offloads mask. It is fully defined at
 *   compile time and may be used for optimization.
 *
 * @return
 *  true - packet match with eMPW batch attributes.
 *  false - no match, eMPW should be restarted.
 */
static __rte_always_inline void
mlx5_tx_idone_empw(struct mlx5_txq_data *restrict txq,
		   struct mlx5_txq_local *restrict loc,
		   unsigned int len,
		   unsigned int slen,
		   unsigned int olx __rte_unused)
{
	assert(MLX5_TXOFF_CONFIG(INLINE));
	assert((len % MLX5_WSEG_SIZE) == 0);
#ifdef MLX5_PMD_SOFT_COUNTERS
	/* Update sent data bytes counter. */
	 txq->stats.obytes += slen;
#else
	(void)slen;
#endif
	len = len / MLX5_WSEG_SIZE + 2;
	loc->wqe_last->cseg.sq_ds = rte_cpu_to_be_32(txq->qp_num_8s | len);
	txq->wqe_ci += (len + 3) / 4;
	loc->wqe_free -= (len + 3) / 4;
	/* Request CQE generation if limits are reached. */
	mlx5_tx_request_completion(txq, loc, false, olx);
}

/**
 * The set of Tx burst functions for single-segment packets
 * without TSO and with Multi-Packet Writing feature support.
 * Supports all types of Tx offloads, except multi-packets
 * and TSO.
 *
 * Uses MLX5_OPCODE_EMPW to build WQEs if possible and sends
 * as many packet per WQE as it can. If eMPW is not configured
 * or packet can not be sent with eMPW (VLAN insertion) the
 * ordinary SEND opcode is used and only one packet placed
 * in WQE.
 *
 * Functions stop sending if it encounters the multi-segment
 * packet or packet with TSO requested.
 *
 * The routines are responsible for storing processed mbuf
 * into elts ring buffer and update elts_head if inlining
 * offload is requested. Otherwise the copying mbufs to elts
 * can be postponed and completed at the end of burst routine.
 *
 * @param txq
 *   Pointer to TX queue structure.
 * @param[in] pkts
 *   Packets to transmit.
 * @param pkts_n
 *   Number of packets in array.
 * @param loc
 *   Pointer to burst routine local context.
 * @param olx
 *   Configured Tx offloads mask. It is fully defined at
 *   compile time and may be used for optimization.
 *
 * @return
 *   MLX5_TXCMP_CODE_EXIT - sending is done or impossible.
 *   MLX5_TXCMP_CODE_ERROR - some unrecoverable error occurred.
 *   MLX5_TXCMP_CODE_MULTI - multi-segment packet encountered.
 *   MLX5_TXCMP_CODE_TSO - TSO packet encountered.
 *   MLX5_TXCMP_CODE_SINGLE - used inside functions set.
 *   MLX5_TXCMP_CODE_EMPW - used inside functions set.
 *
 * Local context variables updated.
 *
 *
 * The routine sends packets with MLX5_OPCODE_EMPW
 * without inlining, this is dedicated optimized branch.
 * No VLAN insertion is supported.
 */
static __rte_always_inline enum mlx5_txcmp_code
mlx5_tx_burst_empw_simple(struct mlx5_txq_data *restrict txq,
			  struct rte_mbuf **restrict pkts,
			  unsigned int pkts_n,
			  struct mlx5_txq_local *restrict loc,
			  unsigned int olx)
{
	/*
	 * Subroutine is the part of mlx5_tx_burst_single()
	 * and sends single-segment packet with eMPW opcode
	 * without data inlining.
	 */
	assert(!MLX5_TXOFF_CONFIG(INLINE));
	assert(MLX5_TXOFF_CONFIG(EMPW));
	assert(loc->elts_free && loc->wqe_free);
	assert(pkts_n > loc->pkts_sent);
	static_assert(MLX5_EMPW_MIN_PACKETS >= 2, "invalid min size");
	pkts += loc->pkts_sent + 1;
	pkts_n -= loc->pkts_sent;
	for (;;) {
		struct mlx5_wqe_dseg *restrict dseg;
		struct mlx5_wqe_eseg *restrict eseg;
		enum mlx5_txcmp_code ret;
		unsigned int part, loop;
		unsigned int slen = 0;

next_empw:
		part = RTE_MIN(pkts_n, MLX5_EMPW_MAX_PACKETS);
		if (unlikely(loc->elts_free < part)) {
			/* We have no enough elts to save all mbufs. */
			if (unlikely(loc->elts_free < MLX5_EMPW_MIN_PACKETS))
				return MLX5_TXCMP_CODE_EXIT;
			/* But we still able to send at least minimal eMPW. */
			part = loc->elts_free;
		}
		/* Check whether we have enough WQEs */
		if (unlikely(loc->wqe_free < ((2 + part + 3) / 4))) {
			if (unlikely(loc->wqe_free <
				((2 + MLX5_EMPW_MIN_PACKETS + 3) / 4)))
				return MLX5_TXCMP_CODE_EXIT;
			part = (loc->wqe_free * 4) - 2;
		}
		if (likely(part > 1))
			rte_prefetch0(*pkts);
		loc->wqe_last = txq->wqes + (txq->wqe_ci & txq->wqe_m);
		/*
		 * Build eMPW title WQEBB:
		 * - Control Segment, eMPW opcode
		 * - Ethernet Segment, no inline
		 */
		mlx5_tx_cseg_init(txq, loc, loc->wqe_last, part + 2,
				  MLX5_OPCODE_ENHANCED_MPSW, olx);
		mlx5_tx_eseg_none(txq, loc, loc->wqe_last,
				  olx & ~MLX5_TXOFF_CONFIG_VLAN);
		eseg = &loc->wqe_last->eseg;
		dseg = &loc->wqe_last->dseg[0];
		loop = part;
		for (;;) {
			uint32_t dlen = rte_pktmbuf_data_len(loc->mbuf);
#ifdef MLX5_PMD_SOFT_COUNTERS
			/* Update sent data bytes counter. */
			slen += dlen;
#endif
			mlx5_tx_dseg_ptr
				(txq, loc, dseg,
				 rte_pktmbuf_mtod(loc->mbuf, uint8_t *),
				 dlen, olx);
			if (unlikely(--loop == 0))
				break;
			loc->mbuf = *pkts++;
			if (likely(loop > 1))
				rte_prefetch0(*pkts);
			ret = mlx5_tx_able_to_empw(txq, loc, olx, true);
			/*
			 * Unroll the completion code to avoid
			 * returning variable value - it results in
			 * unoptimized sequent checking in caller.
			 */
			if (ret == MLX5_TXCMP_CODE_MULTI) {
				part -= loop;
				mlx5_tx_sdone_empw(txq, loc, part, slen, olx);
				if (unlikely(!loc->elts_free ||
					     !loc->wqe_free))
					return MLX5_TXCMP_CODE_EXIT;
				return MLX5_TXCMP_CODE_MULTI;
			}
			if (ret == MLX5_TXCMP_CODE_TSO) {
				part -= loop;
				mlx5_tx_sdone_empw(txq, loc, part, slen, olx);
				if (unlikely(!loc->elts_free ||
					     !loc->wqe_free))
					return MLX5_TXCMP_CODE_EXIT;
				return MLX5_TXCMP_CODE_TSO;
			}
			if (ret == MLX5_TXCMP_CODE_SINGLE) {
				part -= loop;
				mlx5_tx_sdone_empw(txq, loc, part, slen, olx);
				if (unlikely(!loc->elts_free ||
					     !loc->wqe_free))
					return MLX5_TXCMP_CODE_EXIT;
				return MLX5_TXCMP_CODE_SINGLE;
			}
			if (ret != MLX5_TXCMP_CODE_EMPW) {
				assert(false);
				part -= loop;
				mlx5_tx_sdone_empw(txq, loc, part, slen, olx);
				return MLX5_TXCMP_CODE_ERROR;
			}
			/*
			 * Check whether packet parameters coincide
			 * within assumed eMPW batch:
			 * - check sum settings
			 * - metadata value
			 * - software parser settings
			 */
			if (!mlx5_tx_match_empw(txq, eseg, loc, olx)) {
				assert(loop);
				part -= loop;
				mlx5_tx_sdone_empw(txq, loc, part, slen, olx);
				if (unlikely(!loc->elts_free ||
					     !loc->wqe_free))
					return MLX5_TXCMP_CODE_EXIT;
				pkts_n -= part;
				goto next_empw;
			}
			/* Packet attributes match, continue the same eMPW. */
			++dseg;
			if ((uintptr_t)dseg >= (uintptr_t)txq->wqes_end)
				dseg = (struct mlx5_wqe_dseg *)txq->wqes;
		}
		/* eMPW is built successfully, update loop parameters. */
		assert(!loop);
		assert(pkts_n >= part);
#ifdef MLX5_PMD_SOFT_COUNTERS
		/* Update sent data bytes counter. */
		txq->stats.obytes += slen;
#endif
		loc->elts_free -= part;
		loc->pkts_sent += part;
		txq->wqe_ci += (2 + part + 3) / 4;
		loc->wqe_free -= (2 + part + 3) / 4;
		pkts_n -= part;
		/* Request CQE generation if limits are reached. */
		mlx5_tx_request_completion(txq, loc, false, olx);
		if (unlikely(!pkts_n || !loc->elts_free || !loc->wqe_free))
			return MLX5_TXCMP_CODE_EXIT;
		loc->mbuf = *pkts++;
		ret = mlx5_tx_able_to_empw(txq, loc, olx, true);
		if (unlikely(ret != MLX5_TXCMP_CODE_EMPW))
			return ret;
		/* Continue sending eMPW batches. */
	}
	assert(false);
}

/**
 * The routine sends packets with MLX5_OPCODE_EMPW
 * with inlining, optionally supports VLAN insertion.
 */
static __rte_always_inline enum mlx5_txcmp_code
mlx5_tx_burst_empw_inline(struct mlx5_txq_data *restrict txq,
			  struct rte_mbuf **restrict pkts,
			  unsigned int pkts_n,
			  struct mlx5_txq_local *restrict loc,
			  unsigned int olx)
{
	/*
	 * Subroutine is the part of mlx5_tx_burst_single()
	 * and sends single-segment packet with eMPW opcode
	 * with data inlining.
	 */
	assert(MLX5_TXOFF_CONFIG(INLINE));
	assert(MLX5_TXOFF_CONFIG(EMPW));
	assert(loc->elts_free && loc->wqe_free);
	assert(pkts_n > loc->pkts_sent);
	static_assert(MLX5_EMPW_MIN_PACKETS >= 2, "invalid min size");
	pkts += loc->pkts_sent + 1;
	pkts_n -= loc->pkts_sent;
	for (;;) {
		struct mlx5_wqe_dseg *restrict dseg;
		struct mlx5_wqe_eseg *restrict eseg;
		enum mlx5_txcmp_code ret;
		unsigned int room, part, nlim;
		unsigned int slen = 0;

		/*
		 * Limits the amount of packets in one WQE
		 * to improve CQE latency generation.
		 */
		nlim = RTE_MIN(pkts_n, MLX5_EMPW_MAX_PACKETS);
		/* Check whether we have minimal amount WQEs */
		if (unlikely(loc->wqe_free <
			    ((2 + MLX5_EMPW_MIN_PACKETS + 3) / 4)))
			return MLX5_TXCMP_CODE_EXIT;
		if (likely(pkts_n > 1))
			rte_prefetch0(*pkts);
		loc->wqe_last = txq->wqes + (txq->wqe_ci & txq->wqe_m);
		/*
		 * Build eMPW title WQEBB:
		 * - Control Segment, eMPW opcode, zero DS
		 * - Ethernet Segment, no inline
		 */
		mlx5_tx_cseg_init(txq, loc, loc->wqe_last, 0,
				  MLX5_OPCODE_ENHANCED_MPSW, olx);
		mlx5_tx_eseg_none(txq, loc, loc->wqe_last,
				  olx & ~MLX5_TXOFF_CONFIG_VLAN);
		eseg = &loc->wqe_last->eseg;
		dseg = &loc->wqe_last->dseg[0];
		room = RTE_MIN(MLX5_WQE_SIZE_MAX / MLX5_WQE_SIZE,
			       loc->wqe_free) * MLX5_WQE_SIZE -
					MLX5_WQE_CSEG_SIZE -
					MLX5_WQE_ESEG_SIZE;
		/* Build WQE till we have space, packets and resources. */
		part = room;
		for (;;) {
			uint32_t dlen = rte_pktmbuf_data_len(loc->mbuf);
			uint8_t *dptr = rte_pktmbuf_mtod(loc->mbuf, uint8_t *);
			unsigned int tlen;

			assert(room >= MLX5_WQE_DSEG_SIZE);
			assert((room % MLX5_WQE_DSEG_SIZE) == 0);
			assert((uintptr_t)dseg < (uintptr_t)txq->wqes_end);
			/*
			 * Some Tx offloads may cause an error if
			 * packet is not long enough, check against
			 * assumed minimal length.
			 */
			if (unlikely(dlen <= MLX5_ESEG_MIN_INLINE_SIZE)) {
				part -= room;
				if (unlikely(!part))
					return MLX5_TXCMP_CODE_ERROR;
				/*
				 * We have some successfully built
				 * packet Data Segments to send.
				 */
				mlx5_tx_idone_empw(txq, loc, part, slen, olx);
				return MLX5_TXCMP_CODE_ERROR;
			}
			/* Inline or not inline - that's the Question. */
			if (dlen > txq->inlen_empw)
				goto pointer_empw;
			/* Inline entire packet, optional VLAN insertion. */
			tlen = sizeof(dseg->bcount) + dlen;
			if (MLX5_TXOFF_CONFIG(VLAN) &&
			    loc->mbuf->ol_flags & PKT_TX_VLAN_PKT) {
				/*
				 * The packet length must be checked in
				 * mlx5_tx_able_to_empw() and packet
				 * fits into inline length guaranteed.
				 */
				assert((dlen + sizeof(struct rte_vlan_hdr)) <=
					txq->inlen_empw);
				tlen += sizeof(struct rte_vlan_hdr);
				if (room < tlen)
					break;
				dseg = mlx5_tx_dseg_vlan(txq, loc, dseg,
							 dptr, dlen, olx);
#ifdef MLX5_PMD_SOFT_COUNTERS
				/* Update sent data bytes counter. */
				slen +=	sizeof(struct rte_vlan_hdr);
#endif
			} else {
				if (room < tlen)
					break;
				dseg = mlx5_tx_dseg_empw(txq, loc, dseg,
							 dptr, dlen, olx);
			}
			tlen = RTE_ALIGN(tlen, MLX5_WSEG_SIZE);
			assert(room >= tlen);
			room -= tlen;
			/*
			 * Packet data are completely inlined,
			 * free the packet immediately.
			 */
			rte_pktmbuf_free_seg(loc->mbuf);
			goto next_mbuf;
pointer_empw:
			/*
			 * Not inlinable VLAN packets are
			 * proceeded outside of this routine.
			 */
			assert(room >= MLX5_WQE_DSEG_SIZE);
			if (MLX5_TXOFF_CONFIG(VLAN))
				assert(!(loc->mbuf->ol_flags &
					 PKT_TX_VLAN_PKT));
			mlx5_tx_dseg_ptr(txq, loc, dseg, dptr, dlen, olx);
			/* We have to store mbuf in elts.*/
			txq->elts[txq->elts_head++ & txq->elts_m] = loc->mbuf;
			room -= MLX5_WQE_DSEG_SIZE;
			/* Ring buffer wraparound is checked at the loop end.*/
			++dseg;
next_mbuf:
#ifdef MLX5_PMD_SOFT_COUNTERS
			/* Update sent data bytes counter. */
			slen += dlen;
#endif
			loc->pkts_sent++;
			loc->elts_free--;
			pkts_n--;
			if (unlikely(!pkts_n || !loc->elts_free)) {
				/*
				 * We have no resources/packets to
				 * continue build descriptors.
				 */
				part -= room;
				mlx5_tx_idone_empw(txq, loc, part, slen, olx);
				return MLX5_TXCMP_CODE_EXIT;
			}
			loc->mbuf = *pkts++;
			if (likely(pkts_n > 1))
				rte_prefetch0(*pkts);
			ret = mlx5_tx_able_to_empw(txq, loc, olx, true);
			/*
			 * Unroll the completion code to avoid
			 * returning variable value - it results in
			 * unoptimized sequent checking in caller.
			 */
			if (ret == MLX5_TXCMP_CODE_MULTI) {
				part -= room;
				mlx5_tx_idone_empw(txq, loc, part, slen, olx);
				if (unlikely(!loc->elts_free ||
					     !loc->wqe_free))
					return MLX5_TXCMP_CODE_EXIT;
				return MLX5_TXCMP_CODE_MULTI;
			}
			if (ret == MLX5_TXCMP_CODE_TSO) {
				part -= room;
				mlx5_tx_idone_empw(txq, loc, part, slen, olx);
				if (unlikely(!loc->elts_free ||
					     !loc->wqe_free))
					return MLX5_TXCMP_CODE_EXIT;
				return MLX5_TXCMP_CODE_TSO;
			}
			if (ret == MLX5_TXCMP_CODE_SINGLE) {
				part -= room;
				mlx5_tx_idone_empw(txq, loc, part, slen, olx);
				if (unlikely(!loc->elts_free ||
					     !loc->wqe_free))
					return MLX5_TXCMP_CODE_EXIT;
				return MLX5_TXCMP_CODE_SINGLE;
			}
			if (ret != MLX5_TXCMP_CODE_EMPW) {
				assert(false);
				part -= room;
				mlx5_tx_idone_empw(txq, loc, part, slen, olx);
				return MLX5_TXCMP_CODE_ERROR;
			}
			/* Check if we have minimal room left. */
			nlim--;
			if (unlikely(!nlim || room < MLX5_WQE_DSEG_SIZE))
				break;
			/*
			 * Check whether packet parameters coincide
			 * within assumed eMPW batch:
			 * - check sum settings
			 * - metadata value
			 * - software parser settings
			 */
			if (!mlx5_tx_match_empw(txq, eseg, loc, olx))
				break;
			/* Packet attributes match, continue the same eMPW. */
			if ((uintptr_t)dseg >= (uintptr_t)txq->wqes_end)
				dseg = (struct mlx5_wqe_dseg *)txq->wqes;
		}
		/*
		 * We get here to close an existing eMPW
		 * session and start the new one.
		 */
		assert(pkts_n);
		part -= room;
		if (unlikely(!part))
			return MLX5_TXCMP_CODE_EXIT;
		mlx5_tx_idone_empw(txq, loc, part, slen, olx);
		if (unlikely(!loc->elts_free ||
			     !loc->wqe_free))
			return MLX5_TXCMP_CODE_EXIT;
		/* Continue the loop with new eMPW session. */
	}
	assert(false);
}

/**
 * The routine sends packets with ordinary MLX5_OPCODE_SEND.
 * Data inlining and VLAN insertion are supported.
 */
static __rte_always_inline enum mlx5_txcmp_code
mlx5_tx_burst_single_send(struct mlx5_txq_data *restrict txq,
			  struct rte_mbuf **restrict pkts,
			  unsigned int pkts_n,
			  struct mlx5_txq_local *restrict loc,
			  unsigned int olx)
{
	/*
	 * Subroutine is the part of mlx5_tx_burst_single()
	 * and sends single-segment packet with SEND opcode.
	 */
	assert(loc->elts_free && loc->wqe_free);
	assert(pkts_n > loc->pkts_sent);
	pkts += loc->pkts_sent + 1;
	pkts_n -= loc->pkts_sent;
	for (;;) {
		struct mlx5_wqe *restrict wqe;
		enum mlx5_txcmp_code ret;

		assert(NB_SEGS(loc->mbuf) == 1);
		if (MLX5_TXOFF_CONFIG(INLINE)) {
			unsigned int inlen, vlan = 0;

			inlen = rte_pktmbuf_data_len(loc->mbuf);
			if (MLX5_TXOFF_CONFIG(VLAN) &&
			    loc->mbuf->ol_flags & PKT_TX_VLAN_PKT) {
				vlan = sizeof(struct rte_vlan_hdr);
				inlen += vlan;
				static_assert((sizeof(struct rte_vlan_hdr) +
					       sizeof(struct rte_ether_hdr)) ==
					       MLX5_ESEG_MIN_INLINE_SIZE,
					       "invalid min inline data size");
			}
			/*
			 * If inlining is enabled at configuration time
			 * the limit must be not less than minimal size.
			 * Otherwise we would do extra check for data
			 * size to avoid crashes due to length overflow.
			 */
			assert(txq->inlen_send >= MLX5_ESEG_MIN_INLINE_SIZE);
			if (inlen <= txq->inlen_send) {
				unsigned int seg_n, wqe_n;

				rte_prefetch0(rte_pktmbuf_mtod
						(loc->mbuf, uint8_t *));
				/* Check against minimal length. */
				if (inlen <= MLX5_ESEG_MIN_INLINE_SIZE)
					return MLX5_TXCMP_CODE_ERROR;
				/*
				 * Completely inlined packet data WQE:
				 * - Control Segment, SEND opcode
				 * - Ethernet Segment, no VLAN insertion
				 * - Data inlined, VLAN optionally inserted
				 * - Alignment to MLX5_WSEG_SIZE
				 * Have to estimate amount of WQEBBs
				 */
				seg_n = (inlen + 3 * MLX5_WSEG_SIZE -
					 MLX5_ESEG_MIN_INLINE_SIZE +
					 MLX5_WSEG_SIZE - 1) / MLX5_WSEG_SIZE;
				/* Check if there are enough WQEBBs. */
				wqe_n = (seg_n + 3) / 4;
				if (wqe_n > loc->wqe_free)
					return MLX5_TXCMP_CODE_EXIT;
				wqe = txq->wqes + (txq->wqe_ci & txq->wqe_m);
				loc->wqe_last = wqe;
				mlx5_tx_cseg_init(txq, loc, wqe, seg_n,
						  MLX5_OPCODE_SEND, olx);
				mlx5_tx_eseg_data(txq, loc, wqe,
						  vlan, inlen, 0, olx);
				txq->wqe_ci += wqe_n;
				loc->wqe_free -= wqe_n;
				/*
				 * Packet data are completely inlined,
				 * free the packet immediately.
				 */
				rte_pktmbuf_free_seg(loc->mbuf);
			} else if (!MLX5_TXOFF_CONFIG(EMPW) &&
				   txq->inlen_mode) {
				/*
				 * If minimal inlining is requested the eMPW
				 * feature should be disabled due to data is
				 * inlined into Ethernet Segment, which can
				 * not contain inlined data for eMPW due to
				 * segment shared for all packets.
				 */
				struct mlx5_wqe_dseg *restrict dseg;
				unsigned int ds;
				uint8_t *dptr;

				/*
				 * The inline-mode settings require
				 * to inline the specified amount of
				 * data bytes to the Ethernet Segment.
				 * We should check the free space in
				 * WQE ring buffer to inline partially.
				 */
				assert(txq->inlen_send >= txq->inlen_mode);
				assert(inlen > txq->inlen_mode);
				assert(txq->inlen_mode >=
						MLX5_ESEG_MIN_INLINE_SIZE);
				/*
				 * Check whether there are enough free WQEBBs:
				 * - Control Segment
				 * - Ethernet Segment
				 * - First Segment of inlined Ethernet data
				 * - ... data continued ...
				 * - Finishing Data Segment of pointer type
				 */
				ds = (MLX5_WQE_CSEG_SIZE +
				      MLX5_WQE_ESEG_SIZE +
				      MLX5_WQE_DSEG_SIZE +
				      txq->inlen_mode -
				      MLX5_ESEG_MIN_INLINE_SIZE +
				      MLX5_WQE_DSEG_SIZE +
				      MLX5_WSEG_SIZE - 1) / MLX5_WSEG_SIZE;
				if (loc->wqe_free < ((ds + 3) / 4))
					return MLX5_TXCMP_CODE_EXIT;
				/*
				 * Build the ordinary SEND WQE:
				 * - Control Segment
				 * - Ethernet Segment, inline inlen_mode bytes
				 * - Data Segment of pointer type
				 */
				wqe = txq->wqes + (txq->wqe_ci & txq->wqe_m);
				loc->wqe_last = wqe;
				mlx5_tx_cseg_init(txq, loc, wqe, ds,
						  MLX5_OPCODE_SEND, olx);
				dseg = mlx5_tx_eseg_data(txq, loc, wqe, vlan,
							 txq->inlen_mode,
							 0, olx);
				dptr = rte_pktmbuf_mtod(loc->mbuf, uint8_t *) +
				       txq->inlen_mode - vlan;
				inlen -= txq->inlen_mode;
				mlx5_tx_dseg_ptr(txq, loc, dseg,
						 dptr, inlen, olx);
				/*
				 * WQE is built, update the loop parameters
				 * and got to the next packet.
				 */
				txq->wqe_ci += (ds + 3) / 4;
				loc->wqe_free -= (ds + 3) / 4;
				/* We have to store mbuf in elts.*/
				assert(MLX5_TXOFF_CONFIG(INLINE));
				txq->elts[txq->elts_head++ & txq->elts_m] =
						loc->mbuf;
				--loc->elts_free;
			} else {
				uint8_t *dptr;
				unsigned int dlen;

				/*
				 * Partially inlined packet data WQE, we have
				 * some space in title WQEBB, we can fill it
				 * with some packet data. It takes one WQEBB,
				 * it is available, no extra space check:
				 * - Control Segment, SEND opcode
				 * - Ethernet Segment, no VLAN insertion
				 * - MLX5_ESEG_MIN_INLINE_SIZE bytes of Data
				 * - Data Segment, pointer type
				 *
				 * We also get here if VLAN insertion is not
				 * supported by HW, the inline is enabled.
				 */
				wqe = txq->wqes + (txq->wqe_ci & txq->wqe_m);
				loc->wqe_last = wqe;
				mlx5_tx_cseg_init(txq, loc, wqe, 4,
						  MLX5_OPCODE_SEND, olx);
				mlx5_tx_eseg_dmin(txq, loc, wqe, vlan, olx);
				dptr = rte_pktmbuf_mtod(loc->mbuf, uint8_t *) +
				       MLX5_ESEG_MIN_INLINE_SIZE - vlan;
				/*
				 * The length check is performed above, by
				 * comparing with txq->inlen_send. We should
				 * not get overflow here.
				 */
				assert(inlen > MLX5_ESEG_MIN_INLINE_SIZE);
				dlen = inlen - MLX5_ESEG_MIN_INLINE_SIZE;
				mlx5_tx_dseg_ptr(txq, loc, &wqe->dseg[1],
						 dptr, dlen, olx);
				++txq->wqe_ci;
				--loc->wqe_free;
				/* We have to store mbuf in elts.*/
				assert(MLX5_TXOFF_CONFIG(INLINE));
				txq->elts[txq->elts_head++ & txq->elts_m] =
						loc->mbuf;
				--loc->elts_free;
			}
#ifdef MLX5_PMD_SOFT_COUNTERS
			/* Update sent data bytes counter. */
			txq->stats.obytes += vlan +
					rte_pktmbuf_data_len(loc->mbuf);
#endif
		} else {
			/*
			 * No inline at all, it means the CPU cycles saving
			 * is prioritized at configuration, we should not
			 * copy any packet data to WQE.
			 *
			 * SEND WQE, one WQEBB:
			 * - Control Segment, SEND opcode
			 * - Ethernet Segment, optional VLAN, no inline
			 * - Data Segment, pointer type
			 */
			wqe = txq->wqes + (txq->wqe_ci & txq->wqe_m);
			loc->wqe_last = wqe;
			mlx5_tx_cseg_init(txq, loc, wqe, 3,
					  MLX5_OPCODE_SEND, olx);
			mlx5_tx_eseg_none(txq, loc, wqe, olx);
			mlx5_tx_dseg_ptr
				(txq, loc, &wqe->dseg[0],
				 rte_pktmbuf_mtod(loc->mbuf, uint8_t *),
				 rte_pktmbuf_data_len(loc->mbuf), olx);
			++txq->wqe_ci;
			--loc->wqe_free;
			/*
			 * We should not store mbuf pointer in elts
			 * if no inlining is configured, this is done
			 * by calling routine in a batch copy.
			 */
			assert(!MLX5_TXOFF_CONFIG(INLINE));
			--loc->elts_free;
#ifdef MLX5_PMD_SOFT_COUNTERS
			/* Update sent data bytes counter. */
			txq->stats.obytes += rte_pktmbuf_data_len(loc->mbuf);
			if (MLX5_TXOFF_CONFIG(VLAN) &&
			    loc->mbuf->ol_flags & PKT_TX_VLAN_PKT)
				txq->stats.obytes +=
					sizeof(struct rte_vlan_hdr);
#endif
		}
		++loc->pkts_sent;
		--pkts_n;
		/* Request CQE generation if limits are reached. */
		mlx5_tx_request_completion(txq, loc, false, olx);
		if (unlikely(!pkts_n || !loc->elts_free || !loc->wqe_free))
			return MLX5_TXCMP_CODE_EXIT;
		loc->mbuf = *pkts++;
		if (pkts_n > 1)
			rte_prefetch0(*pkts);
		ret = mlx5_tx_able_to_empw(txq, loc, olx, true);
		if (unlikely(ret != MLX5_TXCMP_CODE_SINGLE))
			return ret;
	}
	assert(false);
}

static __rte_always_inline enum mlx5_txcmp_code
mlx5_tx_burst_single(struct mlx5_txq_data *restrict txq,
		     struct rte_mbuf **restrict pkts,
		     unsigned int pkts_n,
		     struct mlx5_txq_local *restrict loc,
		     unsigned int olx)
{
	enum mlx5_txcmp_code ret;

	ret = mlx5_tx_able_to_empw(txq, loc, olx, false);
	if (ret == MLX5_TXCMP_CODE_SINGLE)
		goto ordinary_send;
	assert(ret == MLX5_TXCMP_CODE_EMPW);
	for (;;) {
		/* Optimize for inline/no inline eMPW send. */
		ret = (MLX5_TXOFF_CONFIG(INLINE)) ?
			mlx5_tx_burst_empw_inline
				(txq, pkts, pkts_n, loc, olx) :
			mlx5_tx_burst_empw_simple
				(txq, pkts, pkts_n, loc, olx);
		if (ret != MLX5_TXCMP_CODE_SINGLE)
			return ret;
		/* The resources to send one packet should remain. */
		assert(loc->elts_free && loc->wqe_free);
ordinary_send:
		ret = mlx5_tx_burst_single_send(txq, pkts, pkts_n, loc, olx);
		assert(ret != MLX5_TXCMP_CODE_SINGLE);
		if (ret != MLX5_TXCMP_CODE_EMPW)
			return ret;
		/* The resources to send one packet should remain. */
		assert(loc->elts_free && loc->wqe_free);
	}
}

/**
 * DPDK Tx callback template. This is configured template
 * used to generate routines optimized for specified offload setup.
 * One of this generated functions is chosen at SQ configuration
 * time.
 *
 * @param txq
 *   Generic pointer to TX queue structure.
 * @param[in] pkts
 *   Packets to transmit.
 * @param pkts_n
 *   Number of packets in array.
 * @param olx
 *   Configured offloads mask, presents the bits of MLX5_TXOFF_CONFIG_xxx
 *   values. Should be static to take compile time static configuration
 *   advantages.
 *
 * @return
 *   Number of packets successfully transmitted (<= pkts_n).
 */
static __rte_always_inline uint16_t
mlx5_tx_burst_tmpl(struct mlx5_txq_data *restrict txq,
		   struct rte_mbuf **restrict pkts,
		   uint16_t pkts_n,
		   unsigned int olx)
{
	struct mlx5_txq_local loc;
	enum mlx5_txcmp_code ret;
	unsigned int part;

	assert(txq->elts_s >= (uint16_t)(txq->elts_head - txq->elts_tail));
	assert(txq->wqe_s >= (uint16_t)(txq->wqe_ci - txq->wqe_pi));
	if (unlikely(!pkts_n))
		return 0;
	loc.pkts_sent = 0;
	loc.pkts_copy = 0;
	loc.wqe_last = NULL;

send_loop:
	loc.pkts_loop = loc.pkts_sent;
	/*
	 * Check if there are some CQEs, if any:
	 * - process an encountered errors
	 * - process the completed WQEs
	 * - free related mbufs
	 * - doorbell the NIC about processed CQEs
	 */
	rte_prefetch0(*(pkts + loc.pkts_sent));
	mlx5_tx_handle_completion(txq, olx);
	/*
	 * Calculate the number of available resources - elts and WQEs.
	 * There are two possible different scenarios:
	 * - no data inlining into WQEs, one WQEBB may contains upto
	 *   four packets, in this case elts become scarce resource
	 * - data inlining into WQEs, one packet may require multiple
	 *   WQEBBs, the WQEs become the limiting factor.
	 */
	assert(txq->elts_s >= (uint16_t)(txq->elts_head - txq->elts_tail));
	loc.elts_free = txq->elts_s -
				(uint16_t)(txq->elts_head - txq->elts_tail);
	assert(txq->wqe_s >= (uint16_t)(txq->wqe_ci - txq->wqe_pi));
	loc.wqe_free = txq->wqe_s -
				(uint16_t)(txq->wqe_ci - txq->wqe_pi);
	if (unlikely(!loc.elts_free || !loc.wqe_free))
		return loc.pkts_sent;
	for (;;) {
		/*
		 * Fetch the packet from array. Usually this is
		 * the first packet in series of multi/single
		 * segment packets.
		 */
		loc.mbuf = *(pkts + loc.pkts_sent);
		/* Dedicated branch for multi-segment packets. */
		if (MLX5_TXOFF_CONFIG(MULTI) &&
		    unlikely(NB_SEGS(loc.mbuf) > 1)) {
			/*
			 * Multi-segment packet encountered.
			 * Hardware is able to process it only
			 * with SEND/TSO opcodes, one packet
			 * per WQE, do it in dedicated routine.
			 */
enter_send_multi:
			assert(loc.pkts_sent >= loc.pkts_copy);
			part = loc.pkts_sent - loc.pkts_copy;
			if (!MLX5_TXOFF_CONFIG(INLINE) && part) {
				/*
				 * There are some single-segment mbufs not
				 * stored in elts. The mbufs must be in the
				 * same order as WQEs, so we must copy the
				 * mbufs to elts here, before the coming
				 * multi-segment packet mbufs is appended.
				 */
				mlx5_tx_copy_elts(txq, pkts + loc.pkts_copy,
						  part, olx);
				loc.pkts_copy = loc.pkts_sent;
			}
			assert(pkts_n > loc.pkts_sent);
			ret = mlx5_tx_burst_mseg(txq, pkts, pkts_n, &loc, olx);
			if (!MLX5_TXOFF_CONFIG(INLINE))
				loc.pkts_copy = loc.pkts_sent;
			/*
			 * These returned code checks are supposed
			 * to be optimized out due to routine inlining.
			 */
			if (ret == MLX5_TXCMP_CODE_EXIT) {
				/*
				 * The routine returns this code when
				 * all packets are sent or there is no
				 * enough resources to complete request.
				 */
				break;
			}
			if (ret == MLX5_TXCMP_CODE_ERROR) {
				/*
				 * The routine returns this code when
				 * some error in the incoming packets
				 * format occurred.
				 */
				txq->stats.oerrors++;
				break;
			}
			if (ret == MLX5_TXCMP_CODE_SINGLE) {
				/*
				 * The single-segment packet was encountered
				 * in the array, try to send it with the
				 * best optimized way, possible engaging eMPW.
				 */
				goto enter_send_single;
			}
			if (MLX5_TXOFF_CONFIG(TSO) &&
			    ret == MLX5_TXCMP_CODE_TSO) {
				/*
				 * The single-segment TSO packet was
				 * encountered in the array.
				 */
				goto enter_send_tso;
			}
			/* We must not get here. Something is going wrong. */
			assert(false);
			txq->stats.oerrors++;
			break;
		}
		/* Dedicated branch for single-segment TSO packets. */
		if (MLX5_TXOFF_CONFIG(TSO) &&
		    unlikely(loc.mbuf->ol_flags & PKT_TX_TCP_SEG)) {
			/*
			 * TSO might require special way for inlining
			 * (dedicated parameters) and is sent with
			 * MLX5_OPCODE_TSO opcode only, provide this
			 * in dedicated branch.
			 */
enter_send_tso:
			assert(NB_SEGS(loc.mbuf) == 1);
			assert(pkts_n > loc.pkts_sent);
			ret = mlx5_tx_burst_tso(txq, pkts, pkts_n, &loc, olx);
			/*
			 * These returned code checks are supposed
			 * to be optimized out due to routine inlining.
			 */
			if (ret == MLX5_TXCMP_CODE_EXIT)
				break;
			if (ret == MLX5_TXCMP_CODE_ERROR) {
				txq->stats.oerrors++;
				break;
			}
			if (ret == MLX5_TXCMP_CODE_SINGLE)
				goto enter_send_single;
			if (MLX5_TXOFF_CONFIG(MULTI) &&
			    ret == MLX5_TXCMP_CODE_MULTI) {
				/*
				 * The multi-segment packet was
				 * encountered in the array.
				 */
				goto enter_send_multi;
			}
			/* We must not get here. Something is going wrong. */
			assert(false);
			txq->stats.oerrors++;
			break;
		}
		/*
		 * The dedicated branch for the single-segment packets
		 * without TSO. Often these ones can be sent using
		 * MLX5_OPCODE_EMPW with multiple packets in one WQE.
		 * The routine builds the WQEs till it encounters
		 * the TSO or multi-segment packet (in case if these
		 * offloads are requested at SQ configuration time).
		 */
enter_send_single:
		assert(pkts_n > loc.pkts_sent);
		ret = mlx5_tx_burst_single(txq, pkts, pkts_n, &loc, olx);
		/*
		 * These returned code checks are supposed
		 * to be optimized out due to routine inlining.
		 */
		if (ret == MLX5_TXCMP_CODE_EXIT)
			break;
		if (ret == MLX5_TXCMP_CODE_ERROR) {
			txq->stats.oerrors++;
			break;
		}
		if (MLX5_TXOFF_CONFIG(MULTI) &&
		    ret == MLX5_TXCMP_CODE_MULTI) {
			/*
			 * The multi-segment packet was
			 * encountered in the array.
			 */
			goto enter_send_multi;
		}
		if (MLX5_TXOFF_CONFIG(TSO) &&
		    ret == MLX5_TXCMP_CODE_TSO) {
			/*
			 * The single-segment TSO packet was
			 * encountered in the array.
			 */
			goto enter_send_tso;
		}
		/* We must not get here. Something is going wrong. */
		assert(false);
		txq->stats.oerrors++;
		break;
	}
	/*
	 * Main Tx loop is completed, do the rest:
	 * - set completion request if thresholds are reached
	 * - doorbell the hardware
	 * - copy the rest of mbufs to elts (if any)
	 */
	assert(MLX5_TXOFF_CONFIG(INLINE) || loc.pkts_sent >= loc.pkts_copy);
	/* Take a shortcut if nothing is sent. */
	if (unlikely(loc.pkts_sent == loc.pkts_loop))
		return loc.pkts_sent;
	/*
	 * Ring QP doorbell immediately after WQE building completion
	 * to improve latencies. The pure software related data treatment
	 * can be completed after doorbell. Tx CQEs for this SQ are
	 * processed in this thread only by the polling.
	 */
	mlx5_tx_dbrec_cond_wmb(txq, loc.wqe_last, 0);
	/* Not all of the mbufs may be stored into elts yet. */
	part = MLX5_TXOFF_CONFIG(INLINE) ? 0 : loc.pkts_sent - loc.pkts_copy;
	if (!MLX5_TXOFF_CONFIG(INLINE) && part) {
		/*
		 * There are some single-segment mbufs not stored in elts.
		 * It can be only if the last packet was single-segment.
		 * The copying is gathered into one place due to it is
		 * a good opportunity to optimize that with SIMD.
		 * Unfortunately if inlining is enabled the gaps in
		 * pointer array may happen due to early freeing of the
		 * inlined mbufs.
		 */
		mlx5_tx_copy_elts(txq, pkts + loc.pkts_copy, part, olx);
		loc.pkts_copy = loc.pkts_sent;
	}
#ifdef MLX5_PMD_SOFT_COUNTERS
	/* Increment sent packets counter. */
	txq->stats.opackets += loc.pkts_sent;
#endif
	assert(txq->elts_s >= (uint16_t)(txq->elts_head - txq->elts_tail));
	assert(txq->wqe_s >= (uint16_t)(txq->wqe_ci - txq->wqe_pi));
	if (pkts_n > loc.pkts_sent) {
		/*
		 * If burst size is large there might be no enough CQE
		 * fetched from completion queue and no enough resources
		 * freed to send all the packets.
		 */
		goto send_loop;
	}
	return loc.pkts_sent;
}

/* Generate routines with Enhanced Multi-Packet Write support. */
MLX5_TXOFF_DECL(full_empw,
		MLX5_TXOFF_CONFIG_FULL | MLX5_TXOFF_CONFIG_EMPW)

MLX5_TXOFF_DECL(none_empw,
		MLX5_TXOFF_CONFIG_NONE | MLX5_TXOFF_CONFIG_EMPW)

MLX5_TXOFF_DECL(md_empw,
		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)

MLX5_TXOFF_DECL(mt_empw,
		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)

MLX5_TXOFF_DECL(mtsc_empw,
		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)

MLX5_TXOFF_DECL(mti_empw,
		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
		MLX5_TXOFF_CONFIG_INLINE |
		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)

MLX5_TXOFF_DECL(mtv_empw,
		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
		MLX5_TXOFF_CONFIG_VLAN |
		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)

MLX5_TXOFF_DECL(mtiv_empw,
		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)

MLX5_TXOFF_DECL(sc_empw,
		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)

MLX5_TXOFF_DECL(sci_empw,
		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
		MLX5_TXOFF_CONFIG_INLINE |
		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)

MLX5_TXOFF_DECL(scv_empw,
		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
		MLX5_TXOFF_CONFIG_VLAN |
		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)

MLX5_TXOFF_DECL(sciv_empw,
		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)

MLX5_TXOFF_DECL(i_empw,
		MLX5_TXOFF_CONFIG_INLINE |
		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)

MLX5_TXOFF_DECL(v_empw,
		MLX5_TXOFF_CONFIG_VLAN |
		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)

MLX5_TXOFF_DECL(iv_empw,
		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)

/* Generate routines without Enhanced Multi-Packet Write support. */
MLX5_TXOFF_DECL(full,
		MLX5_TXOFF_CONFIG_FULL)

MLX5_TXOFF_DECL(none,
		MLX5_TXOFF_CONFIG_NONE)

MLX5_TXOFF_DECL(md,
		MLX5_TXOFF_CONFIG_METADATA)

MLX5_TXOFF_DECL(mt,
		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
		MLX5_TXOFF_CONFIG_METADATA)

MLX5_TXOFF_DECL(mtsc,
		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
		MLX5_TXOFF_CONFIG_METADATA)

MLX5_TXOFF_DECL(mti,
		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
		MLX5_TXOFF_CONFIG_INLINE |
		MLX5_TXOFF_CONFIG_METADATA)


MLX5_TXOFF_DECL(mtv,
		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
		MLX5_TXOFF_CONFIG_VLAN |
		MLX5_TXOFF_CONFIG_METADATA)


MLX5_TXOFF_DECL(mtiv,
		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
		MLX5_TXOFF_CONFIG_METADATA)

MLX5_TXOFF_DECL(sc,
		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
		MLX5_TXOFF_CONFIG_METADATA)

MLX5_TXOFF_DECL(sci,
		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
		MLX5_TXOFF_CONFIG_INLINE |
		MLX5_TXOFF_CONFIG_METADATA)


MLX5_TXOFF_DECL(scv,
		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
		MLX5_TXOFF_CONFIG_VLAN |
		MLX5_TXOFF_CONFIG_METADATA)


MLX5_TXOFF_DECL(sciv,
		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
		MLX5_TXOFF_CONFIG_METADATA)

MLX5_TXOFF_DECL(i,
		MLX5_TXOFF_CONFIG_INLINE |
		MLX5_TXOFF_CONFIG_METADATA)

MLX5_TXOFF_DECL(v,
		MLX5_TXOFF_CONFIG_VLAN |
		MLX5_TXOFF_CONFIG_METADATA)

MLX5_TXOFF_DECL(iv,
		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
		MLX5_TXOFF_CONFIG_METADATA)

/*
 * Array of declared and compiled Tx burst function and corresponding
 * supported offloads set. The array is used to select the Tx burst
 * function for specified offloads set at Tx queue configuration time.
 */
const struct {
	eth_tx_burst_t func;
	unsigned int olx;
} txoff_func[] = {
MLX5_TXOFF_INFO(full_empw,
		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)

MLX5_TXOFF_INFO(none_empw,
		MLX5_TXOFF_CONFIG_NONE | MLX5_TXOFF_CONFIG_EMPW)

MLX5_TXOFF_INFO(md_empw,
		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)

MLX5_TXOFF_INFO(mt_empw,
		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)

MLX5_TXOFF_INFO(mtsc_empw,
		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)

MLX5_TXOFF_INFO(mti_empw,
		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
		MLX5_TXOFF_CONFIG_INLINE |
		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)

MLX5_TXOFF_INFO(mtv_empw,
		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
		MLX5_TXOFF_CONFIG_VLAN |
		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)

MLX5_TXOFF_INFO(mtiv_empw,
		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)

MLX5_TXOFF_INFO(sc_empw,
		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)

MLX5_TXOFF_INFO(sci_empw,
		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
		MLX5_TXOFF_CONFIG_INLINE |
		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)

MLX5_TXOFF_INFO(scv_empw,
		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
		MLX5_TXOFF_CONFIG_VLAN |
		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)

MLX5_TXOFF_INFO(sciv_empw,
		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)

MLX5_TXOFF_INFO(i_empw,
		MLX5_TXOFF_CONFIG_INLINE |
		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)

MLX5_TXOFF_INFO(v_empw,
		MLX5_TXOFF_CONFIG_VLAN |
		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)

MLX5_TXOFF_INFO(iv_empw,
		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)

MLX5_TXOFF_INFO(full,
		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
		MLX5_TXOFF_CONFIG_METADATA)

MLX5_TXOFF_INFO(none,
		MLX5_TXOFF_CONFIG_NONE)

MLX5_TXOFF_INFO(md,
		MLX5_TXOFF_CONFIG_METADATA)

MLX5_TXOFF_INFO(mt,
		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
		MLX5_TXOFF_CONFIG_METADATA)

MLX5_TXOFF_INFO(mtsc,
		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
		MLX5_TXOFF_CONFIG_METADATA)

MLX5_TXOFF_INFO(mti,
		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
		MLX5_TXOFF_CONFIG_INLINE |
		MLX5_TXOFF_CONFIG_METADATA)


MLX5_TXOFF_INFO(mtv,
		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
		MLX5_TXOFF_CONFIG_VLAN |
		MLX5_TXOFF_CONFIG_METADATA)

MLX5_TXOFF_INFO(mtiv,
		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
		MLX5_TXOFF_CONFIG_METADATA)

MLX5_TXOFF_INFO(sc,
		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
		MLX5_TXOFF_CONFIG_METADATA)

MLX5_TXOFF_INFO(sci,
		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
		MLX5_TXOFF_CONFIG_INLINE |
		MLX5_TXOFF_CONFIG_METADATA)

MLX5_TXOFF_INFO(scv,
		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
		MLX5_TXOFF_CONFIG_VLAN |
		MLX5_TXOFF_CONFIG_METADATA)

MLX5_TXOFF_INFO(sciv,
		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
		MLX5_TXOFF_CONFIG_METADATA)

MLX5_TXOFF_INFO(i,
		MLX5_TXOFF_CONFIG_INLINE |
		MLX5_TXOFF_CONFIG_METADATA)

MLX5_TXOFF_INFO(v,
		MLX5_TXOFF_CONFIG_VLAN |
		MLX5_TXOFF_CONFIG_METADATA)

MLX5_TXOFF_INFO(iv,
		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
		MLX5_TXOFF_CONFIG_METADATA)
};

/**
 * Configure the Tx function to use. The routine checks configured
 * Tx offloads for the device and selects appropriate Tx burst
 * routine. There are multiple Tx burst routines compiled from
 * the same template in the most optimal way for the dedicated
 * Tx offloads set.
 *
 * @param dev
 *   Pointer to private data structure.
 *
 * @return
 *   Pointer to selected Tx burst function.
 */
eth_tx_burst_t
mlx5_select_tx_function(struct rte_eth_dev *dev)
{
	struct mlx5_priv *priv = dev->data->dev_private;
	struct mlx5_dev_config *config = &priv->config;
	uint64_t tx_offloads = dev->data->dev_conf.txmode.offloads;
	unsigned int diff = 0, olx = 0, i, m;

	static_assert(MLX5_WQE_SIZE_MAX / MLX5_WSEG_SIZE <=
		      MLX5_DSEG_MAX, "invalid WQE max size");
	static_assert(MLX5_WQE_CSEG_SIZE == MLX5_WSEG_SIZE,
		      "invalid WQE Control Segment size");
	static_assert(MLX5_WQE_ESEG_SIZE == MLX5_WSEG_SIZE,
		      "invalid WQE Ethernet Segment size");
	static_assert(MLX5_WQE_DSEG_SIZE == MLX5_WSEG_SIZE,
		      "invalid WQE Data Segment size");
	static_assert(MLX5_WQE_SIZE == 4 * MLX5_WSEG_SIZE,
		      "invalid WQE size");
	assert(priv);
	if (tx_offloads & DEV_TX_OFFLOAD_MULTI_SEGS) {
		/* We should support Multi-Segment Packets. */
		olx |= MLX5_TXOFF_CONFIG_MULTI;
	}
	if (tx_offloads & (DEV_TX_OFFLOAD_TCP_TSO |
			   DEV_TX_OFFLOAD_VXLAN_TNL_TSO |
			   DEV_TX_OFFLOAD_GRE_TNL_TSO |
			   DEV_TX_OFFLOAD_IP_TNL_TSO |
			   DEV_TX_OFFLOAD_UDP_TNL_TSO)) {
		/* We should support TCP Send Offload. */
		olx |= MLX5_TXOFF_CONFIG_TSO;
	}
	if (tx_offloads & (DEV_TX_OFFLOAD_IP_TNL_TSO |
			   DEV_TX_OFFLOAD_UDP_TNL_TSO |
			   DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM)) {
		/* We should support Software Parser for Tunnels. */
		olx |= MLX5_TXOFF_CONFIG_SWP;
	}
	if (tx_offloads & (DEV_TX_OFFLOAD_IPV4_CKSUM |
			   DEV_TX_OFFLOAD_UDP_CKSUM |
			   DEV_TX_OFFLOAD_TCP_CKSUM |
			   DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM)) {
		/* We should support IP/TCP/UDP Checksums. */
		olx |= MLX5_TXOFF_CONFIG_CSUM;
	}
	if (tx_offloads & DEV_TX_OFFLOAD_VLAN_INSERT) {
		/* We should support VLAN insertion. */
		olx |= MLX5_TXOFF_CONFIG_VLAN;
	}
	if (priv->txqs_n && (*priv->txqs)[0]) {
		struct mlx5_txq_data *txd = (*priv->txqs)[0];

		if (txd->inlen_send) {
			/*
			 * Check the data inline requirements. Data inline
			 * is enabled on per device basis, we can check
			 * the first Tx queue only.
			 *
			 * If device does not support VLAN insertion in WQE
			 * and some queues are requested to perform VLAN
			 * insertion offload than inline must be enabled.
			 */
			olx |= MLX5_TXOFF_CONFIG_INLINE;
		}
	}
	if (config->mps == MLX5_MPW_ENHANCED &&
	    config->txq_inline_min <= 0) {
		/*
		 * The NIC supports Enhanced Multi-Packet Write.
		 * We do not support legacy MPW due to its
		 * hardware related problems, so we just ignore
		 * legacy MLX5_MPW settings. There should be no
		 * minimal required inline data.
		 */
		olx |= MLX5_TXOFF_CONFIG_EMPW;
	}
	if (tx_offloads & DEV_TX_OFFLOAD_MATCH_METADATA) {
		/* We should support Flow metadata. */
		olx |= MLX5_TXOFF_CONFIG_METADATA;
	}
	/*
	 * Scan the routines table to find the minimal
	 * satisfying routine with requested offloads.
	 */
	m = RTE_DIM(txoff_func);
	for (i = 0; i < RTE_DIM(txoff_func); i++) {
		unsigned int tmp;

		tmp = txoff_func[i].olx;
		if (tmp == olx) {
			/* Meets requested offloads exactly.*/
			m = i;
			break;
		}
		if ((tmp & olx) != olx) {
			/* Does not meet requested offloads at all. */
			continue;
		}
		if ((olx ^ tmp) & MLX5_TXOFF_CONFIG_EMPW)
			/* Do not enable eMPW if not configured. */
			continue;
		if ((olx ^ tmp) & MLX5_TXOFF_CONFIG_INLINE)
			/* Do not enable inlining if not configured. */
			continue;
		/*
		 * Some routine meets the requirements.
		 * Check whether it has minimal amount
		 * of not requested offloads.
		 */
		tmp = __builtin_popcountl(tmp & ~olx);
		if (m >= RTE_DIM(txoff_func) || tmp < diff) {
			/* First or better match, save and continue. */
			m = i;
			diff = tmp;
			continue;
		}
		if (tmp == diff) {
			tmp = txoff_func[i].olx ^ txoff_func[m].olx;
			if (__builtin_ffsl(txoff_func[i].olx & ~tmp) <
			    __builtin_ffsl(txoff_func[m].olx & ~tmp)) {
				/* Lighter not requested offload. */
				m = i;
			}
		}
	}
	if (m >= RTE_DIM(txoff_func)) {
		DRV_LOG(DEBUG, "port %u has no selected Tx function"
			       " for requested offloads %04X",
				dev->data->port_id, olx);
		return NULL;
	}
	DRV_LOG(DEBUG, "port %u has selected Tx function"
		       " supporting offloads %04X/%04X",
			dev->data->port_id, olx, txoff_func[m].olx);
	if (txoff_func[m].olx & MLX5_TXOFF_CONFIG_MULTI)
		DRV_LOG(DEBUG, "\tMULTI (multi segment)");
	if (txoff_func[m].olx & MLX5_TXOFF_CONFIG_TSO)
		DRV_LOG(DEBUG, "\tTSO   (TCP send offload)");
	if (txoff_func[m].olx & MLX5_TXOFF_CONFIG_SWP)
		DRV_LOG(DEBUG, "\tSWP   (software parser)");
	if (txoff_func[m].olx & MLX5_TXOFF_CONFIG_CSUM)
		DRV_LOG(DEBUG, "\tCSUM  (checksum offload)");
	if (txoff_func[m].olx & MLX5_TXOFF_CONFIG_INLINE)
		DRV_LOG(DEBUG, "\tINLIN (inline data)");
	if (txoff_func[m].olx & MLX5_TXOFF_CONFIG_VLAN)
		DRV_LOG(DEBUG, "\tVLANI (VLAN insertion)");
	if (txoff_func[m].olx & MLX5_TXOFF_CONFIG_METADATA)
		DRV_LOG(DEBUG, "\tMETAD (tx Flow metadata)");
	if (txoff_func[m].olx & MLX5_TXOFF_CONFIG_EMPW)
		DRV_LOG(DEBUG, "\tEMPW  (Enhanced MPW)");
	return txoff_func[m].func;
}