aboutsummaryrefslogtreecommitdiffstats
path: root/drivers/infiniband/hw/hfi1/tid_rdma.c
blob: 5b7905ce688f2f3e37f67827ca57cf851e8c27bd (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
5209
5210
5211
5212
5213
5214
5215
5216
5217
5218
5219
5220
5221
5222
5223
5224
5225
5226
5227
5228
5229
5230
5231
5232
5233
5234
5235
5236
5237
5238
5239
5240
5241
5242
5243
5244
5245
5246
5247
5248
5249
5250
5251
5252
5253
5254
5255
5256
5257
5258
5259
5260
5261
5262
5263
5264
5265
5266
5267
5268
5269
5270
5271
5272
5273
5274
5275
5276
5277
5278
5279
5280
5281
5282
5283
5284
5285
5286
5287
5288
5289
5290
5291
5292
5293
5294
5295
5296
5297
5298
5299
5300
5301
5302
5303
5304
5305
5306
5307
5308
5309
5310
5311
5312
5313
5314
5315
5316
5317
5318
5319
5320
5321
5322
5323
5324
5325
5326
5327
5328
5329
5330
5331
5332
5333
5334
5335
5336
5337
5338
5339
5340
5341
5342
5343
5344
5345
5346
5347
5348
5349
5350
5351
5352
5353
5354
5355
5356
5357
5358
5359
5360
5361
5362
5363
5364
5365
5366
5367
5368
5369
5370
5371
5372
5373
5374
5375
5376
5377
5378
5379
5380
5381
5382
5383
5384
5385
5386
5387
5388
5389
5390
5391
5392
5393
5394
5395
5396
5397
5398
5399
5400
5401
5402
5403
5404
5405
5406
5407
5408
5409
5410
5411
5412
5413
5414
5415
5416
5417
5418
5419
5420
5421
5422
5423
5424
5425
5426
5427
5428
5429
5430
5431
5432
5433
5434
5435
5436
5437
5438
5439
5440
5441
5442
5443
5444
5445
5446
5447
5448
5449
5450
5451
5452
5453
5454
5455
5456
5457
5458
5459
5460
5461
5462
5463
5464
5465
5466
5467
5468
5469
5470
5471
5472
5473
5474
5475
5476
5477
5478
5479
5480
5481
5482
5483
5484
5485
5486
5487
5488
5489
5490
5491
5492
5493
5494
// SPDX-License-Identifier: (GPL-2.0 OR BSD-3-Clause)
/*
 * Copyright(c) 2018 Intel Corporation.
 *
 */

#include "hfi.h"
#include "qp.h"
#include "rc.h"
#include "verbs.h"
#include "tid_rdma.h"
#include "exp_rcv.h"
#include "trace.h"

/**
 * DOC: TID RDMA READ protocol
 *
 * This is an end-to-end protocol at the hfi1 level between two nodes that
 * improves performance by avoiding data copy on the requester side. It
 * converts a qualified RDMA READ request into a TID RDMA READ request on
 * the requester side and thereafter handles the request and response
 * differently. To be qualified, the RDMA READ request should meet the
 * following:
 * -- The total data length should be greater than 256K;
 * -- The total data length should be a multiple of 4K page size;
 * -- Each local scatter-gather entry should be 4K page aligned;
 * -- Each local scatter-gather entry should be a multiple of 4K page size;
 */

#define RCV_TID_FLOW_TABLE_CTRL_FLOW_VALID_SMASK BIT_ULL(32)
#define RCV_TID_FLOW_TABLE_CTRL_HDR_SUPP_EN_SMASK BIT_ULL(33)
#define RCV_TID_FLOW_TABLE_CTRL_KEEP_AFTER_SEQ_ERR_SMASK BIT_ULL(34)
#define RCV_TID_FLOW_TABLE_CTRL_KEEP_ON_GEN_ERR_SMASK BIT_ULL(35)
#define RCV_TID_FLOW_TABLE_STATUS_SEQ_MISMATCH_SMASK BIT_ULL(37)
#define RCV_TID_FLOW_TABLE_STATUS_GEN_MISMATCH_SMASK BIT_ULL(38)

/* Maximum number of packets within a flow generation. */
#define MAX_TID_FLOW_PSN BIT(HFI1_KDETH_BTH_SEQ_SHIFT)

#define GENERATION_MASK 0xFFFFF

static u32 mask_generation(u32 a)
{
	return a & GENERATION_MASK;
}

/* Reserved generation value to set to unused flows for kernel contexts */
#define KERN_GENERATION_RESERVED mask_generation(U32_MAX)

/*
 * J_KEY for kernel contexts when TID RDMA is used.
 * See generate_jkey() in hfi.h for more information.
 */
#define TID_RDMA_JKEY                   32
#define HFI1_KERNEL_MIN_JKEY HFI1_ADMIN_JKEY_RANGE
#define HFI1_KERNEL_MAX_JKEY (2 * HFI1_ADMIN_JKEY_RANGE - 1)

/* Maximum number of segments in flight per QP request. */
#define TID_RDMA_MAX_READ_SEGS_PER_REQ  6
#define TID_RDMA_MAX_WRITE_SEGS_PER_REQ 4
#define MAX_REQ max_t(u16, TID_RDMA_MAX_READ_SEGS_PER_REQ, \
			TID_RDMA_MAX_WRITE_SEGS_PER_REQ)
#define MAX_FLOWS roundup_pow_of_two(MAX_REQ + 1)

#define MAX_EXPECTED_PAGES     (MAX_EXPECTED_BUFFER / PAGE_SIZE)

#define TID_RDMA_DESTQP_FLOW_SHIFT      11
#define TID_RDMA_DESTQP_FLOW_MASK       0x1f

#define TID_OPFN_QP_CTXT_MASK 0xff
#define TID_OPFN_QP_CTXT_SHIFT 56
#define TID_OPFN_QP_KDETH_MASK 0xff
#define TID_OPFN_QP_KDETH_SHIFT 48
#define TID_OPFN_MAX_LEN_MASK 0x7ff
#define TID_OPFN_MAX_LEN_SHIFT 37
#define TID_OPFN_TIMEOUT_MASK 0x1f
#define TID_OPFN_TIMEOUT_SHIFT 32
#define TID_OPFN_RESERVED_MASK 0x3f
#define TID_OPFN_RESERVED_SHIFT 26
#define TID_OPFN_URG_MASK 0x1
#define TID_OPFN_URG_SHIFT 25
#define TID_OPFN_VER_MASK 0x7
#define TID_OPFN_VER_SHIFT 22
#define TID_OPFN_JKEY_MASK 0x3f
#define TID_OPFN_JKEY_SHIFT 16
#define TID_OPFN_MAX_READ_MASK 0x3f
#define TID_OPFN_MAX_READ_SHIFT 10
#define TID_OPFN_MAX_WRITE_MASK 0x3f
#define TID_OPFN_MAX_WRITE_SHIFT 4

/*
 * OPFN TID layout
 *
 * 63               47               31               15
 * NNNNNNNNKKKKKKKK MMMMMMMMMMMTTTTT DDDDDDUVVVJJJJJJ RRRRRRWWWWWWCCCC
 * 3210987654321098 7654321098765432 1098765432109876 5432109876543210
 * N - the context Number
 * K - the Kdeth_qp
 * M - Max_len
 * T - Timeout
 * D - reserveD
 * V - version
 * U - Urg capable
 * J - Jkey
 * R - max_Read
 * W - max_Write
 * C - Capcode
 */

static u32 tid_rdma_flow_wt;

static void tid_rdma_trigger_resume(struct work_struct *work);
static void hfi1_kern_exp_rcv_free_flows(struct tid_rdma_request *req);
static int hfi1_kern_exp_rcv_alloc_flows(struct tid_rdma_request *req,
					 gfp_t gfp);
static void hfi1_init_trdma_req(struct rvt_qp *qp,
				struct tid_rdma_request *req);
static void hfi1_tid_write_alloc_resources(struct rvt_qp *qp, bool intr_ctx);
static void hfi1_tid_timeout(struct timer_list *t);
static void hfi1_add_tid_reap_timer(struct rvt_qp *qp);
static void hfi1_mod_tid_reap_timer(struct rvt_qp *qp);
static void hfi1_mod_tid_retry_timer(struct rvt_qp *qp);
static int hfi1_stop_tid_retry_timer(struct rvt_qp *qp);
static void hfi1_tid_retry_timeout(struct timer_list *t);
static int make_tid_rdma_ack(struct rvt_qp *qp,
			     struct ib_other_headers *ohdr,
			     struct hfi1_pkt_state *ps);
static void hfi1_do_tid_send(struct rvt_qp *qp);
static u32 read_r_next_psn(struct hfi1_devdata *dd, u8 ctxt, u8 fidx);
static void tid_rdma_rcv_err(struct hfi1_packet *packet,
			     struct ib_other_headers *ohdr,
			     struct rvt_qp *qp, u32 psn, int diff, bool fecn);
static void update_r_next_psn_fecn(struct hfi1_packet *packet,
				   struct hfi1_qp_priv *priv,
				   struct hfi1_ctxtdata *rcd,
				   struct tid_rdma_flow *flow,
				   bool fecn);

static u64 tid_rdma_opfn_encode(struct tid_rdma_params *p)
{
	return
		(((u64)p->qp & TID_OPFN_QP_CTXT_MASK) <<
			TID_OPFN_QP_CTXT_SHIFT) |
		((((u64)p->qp >> 16) & TID_OPFN_QP_KDETH_MASK) <<
			TID_OPFN_QP_KDETH_SHIFT) |
		(((u64)((p->max_len >> PAGE_SHIFT) - 1) &
			TID_OPFN_MAX_LEN_MASK) << TID_OPFN_MAX_LEN_SHIFT) |
		(((u64)p->timeout & TID_OPFN_TIMEOUT_MASK) <<
			TID_OPFN_TIMEOUT_SHIFT) |
		(((u64)p->urg & TID_OPFN_URG_MASK) << TID_OPFN_URG_SHIFT) |
		(((u64)p->jkey & TID_OPFN_JKEY_MASK) << TID_OPFN_JKEY_SHIFT) |
		(((u64)p->max_read & TID_OPFN_MAX_READ_MASK) <<
			TID_OPFN_MAX_READ_SHIFT) |
		(((u64)p->max_write & TID_OPFN_MAX_WRITE_MASK) <<
			TID_OPFN_MAX_WRITE_SHIFT);
}

static void tid_rdma_opfn_decode(struct tid_rdma_params *p, u64 data)
{
	p->max_len = (((data >> TID_OPFN_MAX_LEN_SHIFT) &
		TID_OPFN_MAX_LEN_MASK) + 1) << PAGE_SHIFT;
	p->jkey = (data >> TID_OPFN_JKEY_SHIFT) & TID_OPFN_JKEY_MASK;
	p->max_write = (data >> TID_OPFN_MAX_WRITE_SHIFT) &
		TID_OPFN_MAX_WRITE_MASK;
	p->max_read = (data >> TID_OPFN_MAX_READ_SHIFT) &
		TID_OPFN_MAX_READ_MASK;
	p->qp =
		((((data >> TID_OPFN_QP_KDETH_SHIFT) & TID_OPFN_QP_KDETH_MASK)
			<< 16) |
		((data >> TID_OPFN_QP_CTXT_SHIFT) & TID_OPFN_QP_CTXT_MASK));
	p->urg = (data >> TID_OPFN_URG_SHIFT) & TID_OPFN_URG_MASK;
	p->timeout = (data >> TID_OPFN_TIMEOUT_SHIFT) & TID_OPFN_TIMEOUT_MASK;
}

void tid_rdma_opfn_init(struct rvt_qp *qp, struct tid_rdma_params *p)
{
	struct hfi1_qp_priv *priv = qp->priv;

	p->qp = (kdeth_qp << 16) | priv->rcd->ctxt;
	p->max_len = TID_RDMA_MAX_SEGMENT_SIZE;
	p->jkey = priv->rcd->jkey;
	p->max_read = TID_RDMA_MAX_READ_SEGS_PER_REQ;
	p->max_write = TID_RDMA_MAX_WRITE_SEGS_PER_REQ;
	p->timeout = qp->timeout;
	p->urg = is_urg_masked(priv->rcd);
}

bool tid_rdma_conn_req(struct rvt_qp *qp, u64 *data)
{
	struct hfi1_qp_priv *priv = qp->priv;

	*data = tid_rdma_opfn_encode(&priv->tid_rdma.local);
	return true;
}

bool tid_rdma_conn_reply(struct rvt_qp *qp, u64 data)
{
	struct hfi1_qp_priv *priv = qp->priv;
	struct tid_rdma_params *remote, *old;
	bool ret = true;

	old = rcu_dereference_protected(priv->tid_rdma.remote,
					lockdep_is_held(&priv->opfn.lock));
	data &= ~0xfULL;
	/*
	 * If data passed in is zero, return true so as not to continue the
	 * negotiation process
	 */
	if (!data || !HFI1_CAP_IS_KSET(TID_RDMA))
		goto null;
	/*
	 * If kzalloc fails, return false. This will result in:
	 * * at the requester a new OPFN request being generated to retry
	 *   the negotiation
	 * * at the responder, 0 being returned to the requester so as to
	 *   disable TID RDMA at both the requester and the responder
	 */
	remote = kzalloc(sizeof(*remote), GFP_ATOMIC);
	if (!remote) {
		ret = false;
		goto null;
	}

	tid_rdma_opfn_decode(remote, data);
	priv->tid_timer_timeout_jiffies =
		usecs_to_jiffies((((4096UL * (1UL << remote->timeout)) /
				   1000UL) << 3) * 7);
	trace_hfi1_opfn_param(qp, 0, &priv->tid_rdma.local);
	trace_hfi1_opfn_param(qp, 1, remote);
	rcu_assign_pointer(priv->tid_rdma.remote, remote);
	/*
	 * A TID RDMA READ request's segment size is not equal to
	 * remote->max_len only when the request's data length is smaller
	 * than remote->max_len. In that case, there will be only one segment.
	 * Therefore, when priv->pkts_ps is used to calculate req->cur_seg
	 * during retry, it will lead to req->cur_seg = 0, which is exactly
	 * what is expected.
	 */
	priv->pkts_ps = (u16)rvt_div_mtu(qp, remote->max_len);
	priv->timeout_shift = ilog2(priv->pkts_ps - 1) + 1;
	goto free;
null:
	RCU_INIT_POINTER(priv->tid_rdma.remote, NULL);
	priv->timeout_shift = 0;
free:
	if (old)
		kfree_rcu(old, rcu_head);
	return ret;
}

bool tid_rdma_conn_resp(struct rvt_qp *qp, u64 *data)
{
	bool ret;

	ret = tid_rdma_conn_reply(qp, *data);
	*data = 0;
	/*
	 * If tid_rdma_conn_reply() returns error, set *data as 0 to indicate
	 * TID RDMA could not be enabled. This will result in TID RDMA being
	 * disabled at the requester too.
	 */
	if (ret)
		(void)tid_rdma_conn_req(qp, data);
	return ret;
}

void tid_rdma_conn_error(struct rvt_qp *qp)
{
	struct hfi1_qp_priv *priv = qp->priv;
	struct tid_rdma_params *old;

	old = rcu_dereference_protected(priv->tid_rdma.remote,
					lockdep_is_held(&priv->opfn.lock));
	RCU_INIT_POINTER(priv->tid_rdma.remote, NULL);
	if (old)
		kfree_rcu(old, rcu_head);
}

/* This is called at context initialization time */
int hfi1_kern_exp_rcv_init(struct hfi1_ctxtdata *rcd, int reinit)
{
	if (reinit)
		return 0;

	BUILD_BUG_ON(TID_RDMA_JKEY < HFI1_KERNEL_MIN_JKEY);
	BUILD_BUG_ON(TID_RDMA_JKEY > HFI1_KERNEL_MAX_JKEY);
	rcd->jkey = TID_RDMA_JKEY;
	hfi1_set_ctxt_jkey(rcd->dd, rcd, rcd->jkey);
	return hfi1_alloc_ctxt_rcv_groups(rcd);
}

/**
 * qp_to_rcd - determine the receive context used by a qp
 * @qp - the qp
 *
 * This routine returns the receive context associated
 * with a a qp's qpn.
 *
 * Returns the context.
 */
static struct hfi1_ctxtdata *qp_to_rcd(struct rvt_dev_info *rdi,
				       struct rvt_qp *qp)
{
	struct hfi1_ibdev *verbs_dev = container_of(rdi,
						    struct hfi1_ibdev,
						    rdi);
	struct hfi1_devdata *dd = container_of(verbs_dev,
					       struct hfi1_devdata,
					       verbs_dev);
	unsigned int ctxt;

	if (qp->ibqp.qp_num == 0)
		ctxt = 0;
	else
		ctxt = hfi1_get_qp_map(dd, qp->ibqp.qp_num >> dd->qos_shift);
	return dd->rcd[ctxt];
}

int hfi1_qp_priv_init(struct rvt_dev_info *rdi, struct rvt_qp *qp,
		      struct ib_qp_init_attr *init_attr)
{
	struct hfi1_qp_priv *qpriv = qp->priv;
	int i, ret;

	qpriv->rcd = qp_to_rcd(rdi, qp);

	spin_lock_init(&qpriv->opfn.lock);
	INIT_WORK(&qpriv->opfn.opfn_work, opfn_send_conn_request);
	INIT_WORK(&qpriv->tid_rdma.trigger_work, tid_rdma_trigger_resume);
	qpriv->flow_state.psn = 0;
	qpriv->flow_state.index = RXE_NUM_TID_FLOWS;
	qpriv->flow_state.last_index = RXE_NUM_TID_FLOWS;
	qpriv->flow_state.generation = KERN_GENERATION_RESERVED;
	qpriv->s_state = TID_OP(WRITE_RESP);
	qpriv->s_tid_cur = HFI1_QP_WQE_INVALID;
	qpriv->s_tid_head = HFI1_QP_WQE_INVALID;
	qpriv->s_tid_tail = HFI1_QP_WQE_INVALID;
	qpriv->rnr_nak_state = TID_RNR_NAK_INIT;
	qpriv->r_tid_head = HFI1_QP_WQE_INVALID;
	qpriv->r_tid_tail = HFI1_QP_WQE_INVALID;
	qpriv->r_tid_ack = HFI1_QP_WQE_INVALID;
	qpriv->r_tid_alloc = HFI1_QP_WQE_INVALID;
	atomic_set(&qpriv->n_requests, 0);
	atomic_set(&qpriv->n_tid_requests, 0);
	timer_setup(&qpriv->s_tid_timer, hfi1_tid_timeout, 0);
	timer_setup(&qpriv->s_tid_retry_timer, hfi1_tid_retry_timeout, 0);
	INIT_LIST_HEAD(&qpriv->tid_wait);

	if (init_attr->qp_type == IB_QPT_RC && HFI1_CAP_IS_KSET(TID_RDMA)) {
		struct hfi1_devdata *dd = qpriv->rcd->dd;

		qpriv->pages = kzalloc_node(TID_RDMA_MAX_PAGES *
						sizeof(*qpriv->pages),
					    GFP_KERNEL, dd->node);
		if (!qpriv->pages)
			return -ENOMEM;
		for (i = 0; i < qp->s_size; i++) {
			struct hfi1_swqe_priv *priv;
			struct rvt_swqe *wqe = rvt_get_swqe_ptr(qp, i);

			priv = kzalloc_node(sizeof(*priv), GFP_KERNEL,
					    dd->node);
			if (!priv)
				return -ENOMEM;

			hfi1_init_trdma_req(qp, &priv->tid_req);
			priv->tid_req.e.swqe = wqe;
			wqe->priv = priv;
		}
		for (i = 0; i < rvt_max_atomic(rdi); i++) {
			struct hfi1_ack_priv *priv;

			priv = kzalloc_node(sizeof(*priv), GFP_KERNEL,
					    dd->node);
			if (!priv)
				return -ENOMEM;

			hfi1_init_trdma_req(qp, &priv->tid_req);
			priv->tid_req.e.ack = &qp->s_ack_queue[i];

			ret = hfi1_kern_exp_rcv_alloc_flows(&priv->tid_req,
							    GFP_KERNEL);
			if (ret) {
				kfree(priv);
				return ret;
			}
			qp->s_ack_queue[i].priv = priv;
		}
	}

	return 0;
}

void hfi1_qp_priv_tid_free(struct rvt_dev_info *rdi, struct rvt_qp *qp)
{
	struct hfi1_qp_priv *qpriv = qp->priv;
	struct rvt_swqe *wqe;
	u32 i;

	if (qp->ibqp.qp_type == IB_QPT_RC && HFI1_CAP_IS_KSET(TID_RDMA)) {
		for (i = 0; i < qp->s_size; i++) {
			wqe = rvt_get_swqe_ptr(qp, i);
			kfree(wqe->priv);
			wqe->priv = NULL;
		}
		for (i = 0; i < rvt_max_atomic(rdi); i++) {
			struct hfi1_ack_priv *priv = qp->s_ack_queue[i].priv;

			if (priv)
				hfi1_kern_exp_rcv_free_flows(&priv->tid_req);
			kfree(priv);
			qp->s_ack_queue[i].priv = NULL;
		}
		cancel_work_sync(&qpriv->opfn.opfn_work);
		kfree(qpriv->pages);
		qpriv->pages = NULL;
	}
}

/* Flow and tid waiter functions */
/**
 * DOC: lock ordering
 *
 * There are two locks involved with the queuing
 * routines: the qp s_lock and the exp_lock.
 *
 * Since the tid space allocation is called from
 * the send engine, the qp s_lock is already held.
 *
 * The allocation routines will get the exp_lock.
 *
 * The first_qp() call is provided to allow the head of
 * the rcd wait queue to be fetched under the exp_lock and
 * followed by a drop of the exp_lock.
 *
 * Any qp in the wait list will have the qp reference count held
 * to hold the qp in memory.
 */

/*
 * return head of rcd wait list
 *
 * Must hold the exp_lock.
 *
 * Get a reference to the QP to hold the QP in memory.
 *
 * The caller must release the reference when the local
 * is no longer being used.
 */
static struct rvt_qp *first_qp(struct hfi1_ctxtdata *rcd,
			       struct tid_queue *queue)
	__must_hold(&rcd->exp_lock)
{
	struct hfi1_qp_priv *priv;

	lockdep_assert_held(&rcd->exp_lock);
	priv = list_first_entry_or_null(&queue->queue_head,
					struct hfi1_qp_priv,
					tid_wait);
	if (!priv)
		return NULL;
	rvt_get_qp(priv->owner);
	return priv->owner;
}

/**
 * kernel_tid_waiters - determine rcd wait
 * @rcd: the receive context
 * @qp: the head of the qp being processed
 *
 * This routine will return false IFF
 * the list is NULL or the head of the
 * list is the indicated qp.
 *
 * Must hold the qp s_lock and the exp_lock.
 *
 * Return:
 * false if either of the conditions below are statisfied:
 * 1. The list is empty or
 * 2. The indicated qp is at the head of the list and the
 *    HFI1_S_WAIT_TID_SPACE bit is set in qp->s_flags.
 * true is returned otherwise.
 */
static bool kernel_tid_waiters(struct hfi1_ctxtdata *rcd,
			       struct tid_queue *queue, struct rvt_qp *qp)
	__must_hold(&rcd->exp_lock) __must_hold(&qp->s_lock)
{
	struct rvt_qp *fqp;
	bool ret = true;

	lockdep_assert_held(&qp->s_lock);
	lockdep_assert_held(&rcd->exp_lock);
	fqp = first_qp(rcd, queue);
	if (!fqp || (fqp == qp && (qp->s_flags & HFI1_S_WAIT_TID_SPACE)))
		ret = false;
	rvt_put_qp(fqp);
	return ret;
}

/**
 * dequeue_tid_waiter - dequeue the qp from the list
 * @qp - the qp to remove the wait list
 *
 * This routine removes the indicated qp from the
 * wait list if it is there.
 *
 * This should be done after the hardware flow and
 * tid array resources have been allocated.
 *
 * Must hold the qp s_lock and the rcd exp_lock.
 *
 * It assumes the s_lock to protect the s_flags
 * field and to reliably test the HFI1_S_WAIT_TID_SPACE flag.
 */
static void dequeue_tid_waiter(struct hfi1_ctxtdata *rcd,
			       struct tid_queue *queue, struct rvt_qp *qp)
	__must_hold(&rcd->exp_lock) __must_hold(&qp->s_lock)
{
	struct hfi1_qp_priv *priv = qp->priv;

	lockdep_assert_held(&qp->s_lock);
	lockdep_assert_held(&rcd->exp_lock);
	if (list_empty(&priv->tid_wait))
		return;
	list_del_init(&priv->tid_wait);
	qp->s_flags &= ~HFI1_S_WAIT_TID_SPACE;
	queue->dequeue++;
	rvt_put_qp(qp);
}

/**
 * queue_qp_for_tid_wait - suspend QP on tid space
 * @rcd: the receive context
 * @qp: the qp
 *
 * The qp is inserted at the tail of the rcd
 * wait queue and the HFI1_S_WAIT_TID_SPACE s_flag is set.
 *
 * Must hold the qp s_lock and the exp_lock.
 */
static void queue_qp_for_tid_wait(struct hfi1_ctxtdata *rcd,
				  struct tid_queue *queue, struct rvt_qp *qp)
	__must_hold(&rcd->exp_lock) __must_hold(&qp->s_lock)
{
	struct hfi1_qp_priv *priv = qp->priv;

	lockdep_assert_held(&qp->s_lock);
	lockdep_assert_held(&rcd->exp_lock);
	if (list_empty(&priv->tid_wait)) {
		qp->s_flags |= HFI1_S_WAIT_TID_SPACE;
		list_add_tail(&priv->tid_wait, &queue->queue_head);
		priv->tid_enqueue = ++queue->enqueue;
		rcd->dd->verbs_dev.n_tidwait++;
		trace_hfi1_qpsleep(qp, HFI1_S_WAIT_TID_SPACE);
		rvt_get_qp(qp);
	}
}

/**
 * __trigger_tid_waiter - trigger tid waiter
 * @qp: the qp
 *
 * This is a private entrance to schedule the qp
 * assuming the caller is holding the qp->s_lock.
 */
static void __trigger_tid_waiter(struct rvt_qp *qp)
	__must_hold(&qp->s_lock)
{
	lockdep_assert_held(&qp->s_lock);
	if (!(qp->s_flags & HFI1_S_WAIT_TID_SPACE))
		return;
	trace_hfi1_qpwakeup(qp, HFI1_S_WAIT_TID_SPACE);
	hfi1_schedule_send(qp);
}

/**
 * tid_rdma_schedule_tid_wakeup - schedule wakeup for a qp
 * @qp - the qp
 *
 * trigger a schedule or a waiting qp in a deadlock
 * safe manner.  The qp reference is held prior
 * to this call via first_qp().
 *
 * If the qp trigger was already scheduled (!rval)
 * the the reference is dropped, otherwise the resume
 * or the destroy cancel will dispatch the reference.
 */
static void tid_rdma_schedule_tid_wakeup(struct rvt_qp *qp)
{
	struct hfi1_qp_priv *priv;
	struct hfi1_ibport *ibp;
	struct hfi1_pportdata *ppd;
	struct hfi1_devdata *dd;
	bool rval;

	if (!qp)
		return;

	priv = qp->priv;
	ibp = to_iport(qp->ibqp.device, qp->port_num);
	ppd = ppd_from_ibp(ibp);
	dd = dd_from_ibdev(qp->ibqp.device);

	rval = queue_work_on(priv->s_sde ?
			     priv->s_sde->cpu :
			     cpumask_first(cpumask_of_node(dd->node)),
			     ppd->hfi1_wq,
			     &priv->tid_rdma.trigger_work);
	if (!rval)
		rvt_put_qp(qp);
}

/**
 * tid_rdma_trigger_resume - field a trigger work request
 * @work - the work item
 *
 * Complete the off qp trigger processing by directly
 * calling the progress routine.
 */
static void tid_rdma_trigger_resume(struct work_struct *work)
{
	struct tid_rdma_qp_params *tr;
	struct hfi1_qp_priv *priv;
	struct rvt_qp *qp;

	tr = container_of(work, struct tid_rdma_qp_params, trigger_work);
	priv = container_of(tr, struct hfi1_qp_priv, tid_rdma);
	qp = priv->owner;
	spin_lock_irq(&qp->s_lock);
	if (qp->s_flags & HFI1_S_WAIT_TID_SPACE) {
		spin_unlock_irq(&qp->s_lock);
		hfi1_do_send(priv->owner, true);
	} else {
		spin_unlock_irq(&qp->s_lock);
	}
	rvt_put_qp(qp);
}

/**
 * tid_rdma_flush_wait - unwind any tid space wait
 *
 * This is called when resetting a qp to
 * allow a destroy or reset to get rid
 * of any tid space linkage and reference counts.
 */
static void _tid_rdma_flush_wait(struct rvt_qp *qp, struct tid_queue *queue)
	__must_hold(&qp->s_lock)
{
	struct hfi1_qp_priv *priv;

	if (!qp)
		return;
	lockdep_assert_held(&qp->s_lock);
	priv = qp->priv;
	qp->s_flags &= ~HFI1_S_WAIT_TID_SPACE;
	spin_lock(&priv->rcd->exp_lock);
	if (!list_empty(&priv->tid_wait)) {
		list_del_init(&priv->tid_wait);
		qp->s_flags &= ~HFI1_S_WAIT_TID_SPACE;
		queue->dequeue++;
		rvt_put_qp(qp);
	}
	spin_unlock(&priv->rcd->exp_lock);
}

void hfi1_tid_rdma_flush_wait(struct rvt_qp *qp)
	__must_hold(&qp->s_lock)
{
	struct hfi1_qp_priv *priv = qp->priv;

	_tid_rdma_flush_wait(qp, &priv->rcd->flow_queue);
	_tid_rdma_flush_wait(qp, &priv->rcd->rarr_queue);
}

/* Flow functions */
/**
 * kern_reserve_flow - allocate a hardware flow
 * @rcd - the context to use for allocation
 * @last - the index of the preferred flow. Use RXE_NUM_TID_FLOWS to
 *         signify "don't care".
 *
 * Use a bit mask based allocation to reserve a hardware
 * flow for use in receiving KDETH data packets. If a preferred flow is
 * specified the function will attempt to reserve that flow again, if
 * available.
 *
 * The exp_lock must be held.
 *
 * Return:
 * On success: a value postive value between 0 and RXE_NUM_TID_FLOWS - 1
 * On failure: -EAGAIN
 */
static int kern_reserve_flow(struct hfi1_ctxtdata *rcd, int last)
	__must_hold(&rcd->exp_lock)
{
	int nr;

	/* Attempt to reserve the preferred flow index */
	if (last >= 0 && last < RXE_NUM_TID_FLOWS &&
	    !test_and_set_bit(last, &rcd->flow_mask))
		return last;

	nr = ffz(rcd->flow_mask);
	BUILD_BUG_ON(RXE_NUM_TID_FLOWS >=
		     (sizeof(rcd->flow_mask) * BITS_PER_BYTE));
	if (nr > (RXE_NUM_TID_FLOWS - 1))
		return -EAGAIN;
	set_bit(nr, &rcd->flow_mask);
	return nr;
}

static void kern_set_hw_flow(struct hfi1_ctxtdata *rcd, u32 generation,
			     u32 flow_idx)
{
	u64 reg;

	reg = ((u64)generation << HFI1_KDETH_BTH_SEQ_SHIFT) |
		RCV_TID_FLOW_TABLE_CTRL_FLOW_VALID_SMASK |
		RCV_TID_FLOW_TABLE_CTRL_KEEP_AFTER_SEQ_ERR_SMASK |
		RCV_TID_FLOW_TABLE_CTRL_KEEP_ON_GEN_ERR_SMASK |
		RCV_TID_FLOW_TABLE_STATUS_SEQ_MISMATCH_SMASK |
		RCV_TID_FLOW_TABLE_STATUS_GEN_MISMATCH_SMASK;

	if (generation != KERN_GENERATION_RESERVED)
		reg |= RCV_TID_FLOW_TABLE_CTRL_HDR_SUPP_EN_SMASK;

	write_uctxt_csr(rcd->dd, rcd->ctxt,
			RCV_TID_FLOW_TABLE + 8 * flow_idx, reg);
}

static u32 kern_setup_hw_flow(struct hfi1_ctxtdata *rcd, u32 flow_idx)
	__must_hold(&rcd->exp_lock)
{
	u32 generation = rcd->flows[flow_idx].generation;

	kern_set_hw_flow(rcd, generation, flow_idx);
	return generation;
}

static u32 kern_flow_generation_next(u32 gen)
{
	u32 generation = mask_generation(gen + 1);

	if (generation == KERN_GENERATION_RESERVED)
		generation = mask_generation(generation + 1);
	return generation;
}

static void kern_clear_hw_flow(struct hfi1_ctxtdata *rcd, u32 flow_idx)
	__must_hold(&rcd->exp_lock)
{
	rcd->flows[flow_idx].generation =
		kern_flow_generation_next(rcd->flows[flow_idx].generation);
	kern_set_hw_flow(rcd, KERN_GENERATION_RESERVED, flow_idx);
}

int hfi1_kern_setup_hw_flow(struct hfi1_ctxtdata *rcd, struct rvt_qp *qp)
{
	struct hfi1_qp_priv *qpriv = (struct hfi1_qp_priv *)qp->priv;
	struct tid_flow_state *fs = &qpriv->flow_state;
	struct rvt_qp *fqp;
	unsigned long flags;
	int ret = 0;

	/* The QP already has an allocated flow */
	if (fs->index != RXE_NUM_TID_FLOWS)
		return ret;

	spin_lock_irqsave(&rcd->exp_lock, flags);
	if (kernel_tid_waiters(rcd, &rcd->flow_queue, qp))
		goto queue;

	ret = kern_reserve_flow(rcd, fs->last_index);
	if (ret < 0)
		goto queue;
	fs->index = ret;
	fs->last_index = fs->index;

	/* Generation received in a RESYNC overrides default flow generation */
	if (fs->generation != KERN_GENERATION_RESERVED)
		rcd->flows[fs->index].generation = fs->generation;
	fs->generation = kern_setup_hw_flow(rcd, fs->index);
	fs->psn = 0;
	dequeue_tid_waiter(rcd, &rcd->flow_queue, qp);
	/* get head before dropping lock */
	fqp = first_qp(rcd, &rcd->flow_queue);
	spin_unlock_irqrestore(&rcd->exp_lock, flags);

	tid_rdma_schedule_tid_wakeup(fqp);
	return 0;
queue:
	queue_qp_for_tid_wait(rcd, &rcd->flow_queue, qp);
	spin_unlock_irqrestore(&rcd->exp_lock, flags);
	return -EAGAIN;
}

void hfi1_kern_clear_hw_flow(struct hfi1_ctxtdata *rcd, struct rvt_qp *qp)
{
	struct hfi1_qp_priv *qpriv = (struct hfi1_qp_priv *)qp->priv;
	struct tid_flow_state *fs = &qpriv->flow_state;
	struct rvt_qp *fqp;
	unsigned long flags;

	if (fs->index >= RXE_NUM_TID_FLOWS)
		return;
	spin_lock_irqsave(&rcd->exp_lock, flags);
	kern_clear_hw_flow(rcd, fs->index);
	clear_bit(fs->index, &rcd->flow_mask);
	fs->index = RXE_NUM_TID_FLOWS;
	fs->psn = 0;
	fs->generation = KERN_GENERATION_RESERVED;

	/* get head before dropping lock */
	fqp = first_qp(rcd, &rcd->flow_queue);
	spin_unlock_irqrestore(&rcd->exp_lock, flags);

	if (fqp == qp) {
		__trigger_tid_waiter(fqp);
		rvt_put_qp(fqp);
	} else {
		tid_rdma_schedule_tid_wakeup(fqp);
	}
}

void hfi1_kern_init_ctxt_generations(struct hfi1_ctxtdata *rcd)
{
	int i;

	for (i = 0; i < RXE_NUM_TID_FLOWS; i++) {
		rcd->flows[i].generation = mask_generation(prandom_u32());
		kern_set_hw_flow(rcd, KERN_GENERATION_RESERVED, i);
	}
}

/* TID allocation functions */
static u8 trdma_pset_order(struct tid_rdma_pageset *s)
{
	u8 count = s->count;

	return ilog2(count) + 1;
}

/**
 * tid_rdma_find_phys_blocks_4k - get groups base on mr info
 * @npages - number of pages
 * @pages - pointer to an array of page structs
 * @list - page set array to return
 *
 * This routine returns the number of groups associated with
 * the current sge information.  This implementation is based
 * on the expected receive find_phys_blocks() adjusted to
 * use the MR information vs. the pfn.
 *
 * Return:
 * the number of RcvArray entries
 */
static u32 tid_rdma_find_phys_blocks_4k(struct tid_rdma_flow *flow,
					struct page **pages,
					u32 npages,
					struct tid_rdma_pageset *list)
{
	u32 pagecount, pageidx, setcount = 0, i;
	void *vaddr, *this_vaddr;

	if (!npages)
		return 0;

	/*
	 * Look for sets of physically contiguous pages in the user buffer.
	 * This will allow us to optimize Expected RcvArray entry usage by
	 * using the bigger supported sizes.
	 */
	vaddr = page_address(pages[0]);
	trace_hfi1_tid_flow_page(flow->req->qp, flow, 0, 0, 0, vaddr);
	for (pageidx = 0, pagecount = 1, i = 1; i <= npages; i++) {
		this_vaddr = i < npages ? page_address(pages[i]) : NULL;
		trace_hfi1_tid_flow_page(flow->req->qp, flow, i, 0, 0,
					 this_vaddr);
		/*
		 * If the vaddr's are not sequential, pages are not physically
		 * contiguous.
		 */
		if (this_vaddr != (vaddr + PAGE_SIZE)) {
			/*
			 * At this point we have to loop over the set of
			 * physically contiguous pages and break them down it
			 * sizes supported by the HW.
			 * There are two main constraints:
			 *     1. The max buffer size is MAX_EXPECTED_BUFFER.
			 *        If the total set size is bigger than that
			 *        program only a MAX_EXPECTED_BUFFER chunk.
			 *     2. The buffer size has to be a power of two. If
			 *        it is not, round down to the closes power of
			 *        2 and program that size.
			 */
			while (pagecount) {
				int maxpages = pagecount;
				u32 bufsize = pagecount * PAGE_SIZE;

				if (bufsize > MAX_EXPECTED_BUFFER)
					maxpages =
						MAX_EXPECTED_BUFFER >>
						PAGE_SHIFT;
				else if (!is_power_of_2(bufsize))
					maxpages =
						rounddown_pow_of_two(bufsize) >>
						PAGE_SHIFT;

				list[setcount].idx = pageidx;
				list[setcount].count = maxpages;
				trace_hfi1_tid_pageset(flow->req->qp, setcount,
						       list[setcount].idx,
						       list[setcount].count);
				pagecount -= maxpages;
				pageidx += maxpages;
				setcount++;
			}
			pageidx = i;
			pagecount = 1;
			vaddr = this_vaddr;
		} else {
			vaddr += PAGE_SIZE;
			pagecount++;
		}
	}
	/* insure we always return an even number of sets */
	if (setcount & 1)
		list[setcount++].count = 0;
	return setcount;
}

/**
 * tid_flush_pages - dump out pages into pagesets
 * @list - list of pagesets
 * @idx - pointer to current page index
 * @pages - number of pages to dump
 * @sets - current number of pagesset
 *
 * This routine flushes out accumuated pages.
 *
 * To insure an even number of sets the
 * code may add a filler.
 *
 * This can happen with when pages is not
 * a power of 2 or pages is a power of 2
 * less than the maximum pages.
 *
 * Return:
 * The new number of sets
 */

static u32 tid_flush_pages(struct tid_rdma_pageset *list,
			   u32 *idx, u32 pages, u32 sets)
{
	while (pages) {
		u32 maxpages = pages;

		if (maxpages > MAX_EXPECTED_PAGES)
			maxpages = MAX_EXPECTED_PAGES;
		else if (!is_power_of_2(maxpages))
			maxpages = rounddown_pow_of_two(maxpages);
		list[sets].idx = *idx;
		list[sets++].count = maxpages;
		*idx += maxpages;
		pages -= maxpages;
	}
	/* might need a filler */
	if (sets & 1)
		list[sets++].count = 0;
	return sets;
}

/**
 * tid_rdma_find_phys_blocks_8k - get groups base on mr info
 * @pages - pointer to an array of page structs
 * @npages - number of pages
 * @list - page set array to return
 *
 * This routine parses an array of pages to compute pagesets
 * in an 8k compatible way.
 *
 * pages are tested two at a time, i, i + 1 for contiguous
 * pages and i - 1 and i contiguous pages.
 *
 * If any condition is false, any accumlated pages are flushed and
 * v0,v1 are emitted as separate PAGE_SIZE pagesets
 *
 * Otherwise, the current 8k is totaled for a future flush.
 *
 * Return:
 * The number of pagesets
 * list set with the returned number of pagesets
 *
 */
static u32 tid_rdma_find_phys_blocks_8k(struct tid_rdma_flow *flow,
					struct page **pages,
					u32 npages,
					struct tid_rdma_pageset *list)
{
	u32 idx, sets = 0, i;
	u32 pagecnt = 0;
	void *v0, *v1, *vm1;

	if (!npages)
		return 0;
	for (idx = 0, i = 0, vm1 = NULL; i < npages; i += 2) {
		/* get a new v0 */
		v0 = page_address(pages[i]);
		trace_hfi1_tid_flow_page(flow->req->qp, flow, i, 1, 0, v0);
		v1 = i + 1 < npages ?
				page_address(pages[i + 1]) : NULL;
		trace_hfi1_tid_flow_page(flow->req->qp, flow, i, 1, 1, v1);
		/* compare i, i + 1 vaddr */
		if (v1 != (v0 + PAGE_SIZE)) {
			/* flush out pages */
			sets = tid_flush_pages(list, &idx, pagecnt, sets);
			/* output v0,v1 as two pagesets */
			list[sets].idx = idx++;
			list[sets++].count = 1;
			if (v1) {
				list[sets].count = 1;
				list[sets++].idx = idx++;
			} else {
				list[sets++].count = 0;
			}
			vm1 = NULL;
			pagecnt = 0;
			continue;
		}
		/* i,i+1 consecutive, look at i-1,i */
		if (vm1 && v0 != (vm1 + PAGE_SIZE)) {
			/* flush out pages */
			sets = tid_flush_pages(list, &idx, pagecnt, sets);
			pagecnt = 0;
		}
		/* pages will always be a multiple of 8k */
		pagecnt += 2;
		/* save i-1 */
		vm1 = v1;
		/* move to next pair */
	}
	/* dump residual pages at end */
	sets = tid_flush_pages(list, &idx, npages - idx, sets);
	/* by design cannot be odd sets */
	WARN_ON(sets & 1);
	return sets;
}

/**
 * Find pages for one segment of a sge array represented by @ss. The function
 * does not check the sge, the sge must have been checked for alignment with a
 * prior call to hfi1_kern_trdma_ok. Other sge checking is done as part of
 * rvt_lkey_ok and rvt_rkey_ok. Also, the function only modifies the local sge
 * copy maintained in @ss->sge, the original sge is not modified.
 *
 * Unlike IB RDMA WRITE, we can't decrement ss->num_sge here because we are not
 * releasing the MR reference count at the same time. Otherwise, we'll "leak"
 * references to the MR. This difference requires that we keep track of progress
 * into the sg_list. This is done by the cur_seg cursor in the tid_rdma_request
 * structure.
 */
static u32 kern_find_pages(struct tid_rdma_flow *flow,
			   struct page **pages,
			   struct rvt_sge_state *ss, bool *last)
{
	struct tid_rdma_request *req = flow->req;
	struct rvt_sge *sge = &ss->sge;
	u32 length = flow->req->seg_len;
	u32 len = PAGE_SIZE;
	u32 i = 0;

	while (length && req->isge < ss->num_sge) {
		pages[i++] = virt_to_page(sge->vaddr);

		sge->vaddr += len;
		sge->length -= len;
		sge->sge_length -= len;
		if (!sge->sge_length) {
			if (++req->isge < ss->num_sge)
				*sge = ss->sg_list[req->isge - 1];
		} else if (sge->length == 0 && sge->mr->lkey) {
			if (++sge->n >= RVT_SEGSZ) {
				++sge->m;
				sge->n = 0;
			}
			sge->vaddr = sge->mr->map[sge->m]->segs[sge->n].vaddr;
			sge->length = sge->mr->map[sge->m]->segs[sge->n].length;
		}
		length -= len;
	}

	flow->length = flow->req->seg_len - length;
	*last = req->isge == ss->num_sge ? false : true;
	return i;
}

static void dma_unmap_flow(struct tid_rdma_flow *flow)
{
	struct hfi1_devdata *dd;
	int i;
	struct tid_rdma_pageset *pset;

	dd = flow->req->rcd->dd;
	for (i = 0, pset = &flow->pagesets[0]; i < flow->npagesets;
			i++, pset++) {
		if (pset->count && pset->addr) {
			dma_unmap_page(&dd->pcidev->dev,
				       pset->addr,
				       PAGE_SIZE * pset->count,
				       DMA_FROM_DEVICE);
			pset->mapped = 0;
		}
	}
}

static int dma_map_flow(struct tid_rdma_flow *flow, struct page **pages)
{
	int i;
	struct hfi1_devdata *dd = flow->req->rcd->dd;
	struct tid_rdma_pageset *pset;

	for (i = 0, pset = &flow->pagesets[0]; i < flow->npagesets;
			i++, pset++) {
		if (pset->count) {
			pset->addr = dma_map_page(&dd->pcidev->dev,
						  pages[pset->idx],
						  0,
						  PAGE_SIZE * pset->count,
						  DMA_FROM_DEVICE);

			if (dma_mapping_error(&dd->pcidev->dev, pset->addr)) {
				dma_unmap_flow(flow);
				return -ENOMEM;
			}
			pset->mapped = 1;
		}
	}
	return 0;
}

static inline bool dma_mapped(struct tid_rdma_flow *flow)
{
	return !!flow->pagesets[0].mapped;
}

/*
 * Get pages pointers and identify contiguous physical memory chunks for a
 * segment. All segments are of length flow->req->seg_len.
 */
static int kern_get_phys_blocks(struct tid_rdma_flow *flow,
				struct page **pages,
				struct rvt_sge_state *ss, bool *last)
{
	u8 npages;

	/* Reuse previously computed pagesets, if any */
	if (flow->npagesets) {
		trace_hfi1_tid_flow_alloc(flow->req->qp, flow->req->setup_head,
					  flow);
		if (!dma_mapped(flow))
			return dma_map_flow(flow, pages);
		return 0;
	}

	npages = kern_find_pages(flow, pages, ss, last);

	if (flow->req->qp->pmtu == enum_to_mtu(OPA_MTU_4096))
		flow->npagesets =
			tid_rdma_find_phys_blocks_4k(flow, pages, npages,
						     flow->pagesets);
	else
		flow->npagesets =
			tid_rdma_find_phys_blocks_8k(flow, pages, npages,
						     flow->pagesets);

	return dma_map_flow(flow, pages);
}

static inline void kern_add_tid_node(struct tid_rdma_flow *flow,
				     struct hfi1_ctxtdata *rcd, char *s,
				     struct tid_group *grp, u8 cnt)
{
	struct kern_tid_node *node = &flow->tnode[flow->tnode_cnt++];

	WARN_ON_ONCE(flow->tnode_cnt >=
		     (TID_RDMA_MAX_SEGMENT_SIZE >> PAGE_SHIFT));
	if (WARN_ON_ONCE(cnt & 1))
		dd_dev_err(rcd->dd,
			   "unexpected odd allocation cnt %u map 0x%x used %u",
			   cnt, grp->map, grp->used);

	node->grp = grp;
	node->map = grp->map;
	node->cnt = cnt;
	trace_hfi1_tid_node_add(flow->req->qp, s, flow->tnode_cnt - 1,
				grp->base, grp->map, grp->used, cnt);
}

/*
 * Try to allocate pageset_count TID's from TID groups for a context
 *
 * This function allocates TID's without moving groups between lists or
 * modifying grp->map. This is done as follows, being cogizant of the lists
 * between which the TID groups will move:
 * 1. First allocate complete groups of 8 TID's since this is more efficient,
 *    these groups will move from group->full without affecting used
 * 2. If more TID's are needed allocate from used (will move from used->full or
 *    stay in used)
 * 3. If we still don't have the required number of TID's go back and look again
 *    at a complete group (will move from group->used)
 */
static int kern_alloc_tids(struct tid_rdma_flow *flow)
{
	struct hfi1_ctxtdata *rcd = flow->req->rcd;
	struct hfi1_devdata *dd = rcd->dd;
	u32 ngroups, pageidx = 0;
	struct tid_group *group = NULL, *used;
	u8 use;

	flow->tnode_cnt = 0;
	ngroups = flow->npagesets / dd->rcv_entries.group_size;
	if (!ngroups)
		goto used_list;

	/* First look at complete groups */
	list_for_each_entry(group,  &rcd->tid_group_list.list, list) {
		kern_add_tid_node(flow, rcd, "complete groups", group,
				  group->size);

		pageidx += group->size;
		if (!--ngroups)
			break;
	}

	if (pageidx >= flow->npagesets)
		goto ok;

used_list:
	/* Now look at partially used groups */
	list_for_each_entry(used, &rcd->tid_used_list.list, list) {
		use = min_t(u32, flow->npagesets - pageidx,
			    used->size - used->used);
		kern_add_tid_node(flow, rcd, "used groups", used, use);

		pageidx += use;
		if (pageidx >= flow->npagesets)
			goto ok;
	}

	/*
	 * Look again at a complete group, continuing from where we left.
	 * However, if we are at the head, we have reached the end of the
	 * complete groups list from the first loop above
	 */
	if (group && &group->list == &rcd->tid_group_list.list)
		goto bail_eagain;
	group = list_prepare_entry(group, &rcd->tid_group_list.list,
				   list);
	if (list_is_last(&group->list, &rcd->tid_group_list.list))
		goto bail_eagain;
	group = list_next_entry(group, list);
	use = min_t(u32, flow->npagesets - pageidx, group->size);
	kern_add_tid_node(flow, rcd, "complete continue", group, use);
	pageidx += use;
	if (pageidx >= flow->npagesets)
		goto ok;
bail_eagain:
	trace_hfi1_msg_alloc_tids(flow->req->qp, " insufficient tids: needed ",
				  (u64)flow->npagesets);
	return -EAGAIN;
ok:
	return 0;
}

static void kern_program_rcv_group(struct tid_rdma_flow *flow, int grp_num,
				   u32 *pset_idx)
{
	struct hfi1_ctxtdata *rcd = flow->req->rcd;
	struct hfi1_devdata *dd = rcd->dd;
	struct kern_tid_node *node = &flow->tnode[grp_num];
	struct tid_group *grp = node->grp;
	struct tid_rdma_pageset *pset;
	u32 pmtu_pg = flow->req->qp->pmtu >> PAGE_SHIFT;
	u32 rcventry, npages = 0, pair = 0, tidctrl;
	u8 i, cnt = 0;

	for (i = 0; i < grp->size; i++) {
		rcventry = grp->base + i;

		if (node->map & BIT(i) || cnt >= node->cnt) {
			rcv_array_wc_fill(dd, rcventry);
			continue;
		}
		pset = &flow->pagesets[(*pset_idx)++];
		if (pset->count) {
			hfi1_put_tid(dd, rcventry, PT_EXPECTED,
				     pset->addr, trdma_pset_order(pset));
		} else {
			hfi1_put_tid(dd, rcventry, PT_INVALID, 0, 0);
		}
		npages += pset->count;

		rcventry -= rcd->expected_base;
		tidctrl = pair ? 0x3 : rcventry & 0x1 ? 0x2 : 0x1;
		/*
		 * A single TID entry will be used to use a rcvarr pair (with
		 * tidctrl 0x3), if ALL these are true (a) the bit pos is even
		 * (b) the group map shows current and the next bits as free
		 * indicating two consecutive rcvarry entries are available (c)
		 * we actually need 2 more entries
		 */
		pair = !(i & 0x1) && !((node->map >> i) & 0x3) &&
			node->cnt >= cnt + 2;
		if (!pair) {
			if (!pset->count)
				tidctrl = 0x1;
			flow->tid_entry[flow->tidcnt++] =
				EXP_TID_SET(IDX, rcventry >> 1) |
				EXP_TID_SET(CTRL, tidctrl) |
				EXP_TID_SET(LEN, npages);
			trace_hfi1_tid_entry_alloc(/* entry */
			   flow->req->qp, flow->tidcnt - 1,
			   flow->tid_entry[flow->tidcnt - 1]);

			/* Efficient DIV_ROUND_UP(npages, pmtu_pg) */
			flow->npkts += (npages + pmtu_pg - 1) >> ilog2(pmtu_pg);
			npages = 0;
		}

		if (grp->used == grp->size - 1)
			tid_group_move(grp, &rcd->tid_used_list,
				       &rcd->tid_full_list);
		else if (!grp->used)
			tid_group_move(grp, &rcd->tid_group_list,
				       &rcd->tid_used_list);

		grp->used++;
		grp->map |= BIT(i);
		cnt++;
	}
}

static void kern_unprogram_rcv_group(struct tid_rdma_flow *flow, int grp_num)
{
	struct hfi1_ctxtdata *rcd = flow->req->rcd;
	struct hfi1_devdata *dd = rcd->dd;
	struct kern_tid_node *node = &flow->tnode[grp_num];
	struct tid_group *grp = node->grp;
	u32 rcventry;
	u8 i, cnt = 0;

	for (i = 0; i < grp->size; i++) {
		rcventry = grp->base + i;

		if (node->map & BIT(i) || cnt >= node->cnt) {
			rcv_array_wc_fill(dd, rcventry);
			continue;
		}

		hfi1_put_tid(dd, rcventry, PT_INVALID, 0, 0);

		grp->used--;
		grp->map &= ~BIT(i);
		cnt++;

		if (grp->used == grp->size - 1)
			tid_group_move(grp, &rcd->tid_full_list,
				       &rcd->tid_used_list);
		else if (!grp->used)
			tid_group_move(grp, &rcd->tid_used_list,
				       &rcd->tid_group_list);
	}
	if (WARN_ON_ONCE(cnt & 1)) {
		struct hfi1_ctxtdata *rcd = flow->req->rcd;
		struct hfi1_devdata *dd = rcd->dd;

		dd_dev_err(dd, "unexpected odd free cnt %u map 0x%x used %u",
			   cnt, grp->map, grp->used);
	}
}

static void kern_program_rcvarray(struct tid_rdma_flow *flow)
{
	u32 pset_idx = 0;
	int i;

	flow->npkts = 0;
	flow->tidcnt = 0;
	for (i = 0; i < flow->tnode_cnt; i++)
		kern_program_rcv_group(flow, i, &pset_idx);
	trace_hfi1_tid_flow_alloc(flow->req->qp, flow->req->setup_head, flow);
}

/**
 * hfi1_kern_exp_rcv_setup() - setup TID's and flow for one segment of a
 * TID RDMA request
 *
 * @req: TID RDMA request for which the segment/flow is being set up
 * @ss: sge state, maintains state across successive segments of a sge
 * @last: set to true after the last sge segment has been processed
 *
 * This function
 * (1) finds a free flow entry in the flow circular buffer
 * (2) finds pages and continuous physical chunks constituing one segment
 *     of an sge
 * (3) allocates TID group entries for those chunks
 * (4) programs rcvarray entries in the hardware corresponding to those
 *     TID's
 * (5) computes a tidarray with formatted TID entries which can be sent
 *     to the sender
 * (6) Reserves and programs HW flows.
 * (7) It also manages queing the QP when TID/flow resources are not
 *     available.
 *
 * @req points to struct tid_rdma_request of which the segments are a part. The
 * function uses qp, rcd and seg_len members of @req. In the absence of errors,
 * req->flow_idx is the index of the flow which has been prepared in this
 * invocation of function call. With flow = &req->flows[req->flow_idx],
 * flow->tid_entry contains the TID array which the sender can use for TID RDMA
 * sends and flow->npkts contains number of packets required to send the
 * segment.
 *
 * hfi1_check_sge_align should be called prior to calling this function and if
 * it signals error TID RDMA cannot be used for this sge and this function
 * should not be called.
 *
 * For the queuing, caller must hold the flow->req->qp s_lock from the send
 * engine and the function will procure the exp_lock.
 *
 * Return:
 * The function returns -EAGAIN if sufficient number of TID/flow resources to
 * map the segment could not be allocated. In this case the function should be
 * called again with previous arguments to retry the TID allocation. There are
 * no other error returns. The function returns 0 on success.
 */
int hfi1_kern_exp_rcv_setup(struct tid_rdma_request *req,
			    struct rvt_sge_state *ss, bool *last)
	__must_hold(&req->qp->s_lock)
{
	struct tid_rdma_flow *flow = &req->flows[req->setup_head];
	struct hfi1_ctxtdata *rcd = req->rcd;
	struct hfi1_qp_priv *qpriv = req->qp->priv;
	unsigned long flags;
	struct rvt_qp *fqp;
	u16 clear_tail = req->clear_tail;

	lockdep_assert_held(&req->qp->s_lock);
	/*
	 * We return error if either (a) we don't have space in the flow
	 * circular buffer, or (b) we already have max entries in the buffer.
	 * Max entries depend on the type of request we are processing and the
	 * negotiated TID RDMA parameters.
	 */
	if (!CIRC_SPACE(req->setup_head, clear_tail, MAX_FLOWS) ||
	    CIRC_CNT(req->setup_head, clear_tail, MAX_FLOWS) >=
	    req->n_flows)
		return -EINVAL;

	/*
	 * Get pages, identify contiguous physical memory chunks for the segment
	 * If we can not determine a DMA address mapping we will treat it just
	 * like if we ran out of space above.
	 */
	if (kern_get_phys_blocks(flow, qpriv->pages, ss, last)) {
		hfi1_wait_kmem(flow->req->qp);
		return -ENOMEM;
	}

	spin_lock_irqsave(&rcd->exp_lock, flags);
	if (kernel_tid_waiters(rcd, &rcd->rarr_queue, flow->req->qp))
		goto queue;

	/*
	 * At this point we know the number of pagesets and hence the number of
	 * TID's to map the segment. Allocate the TID's from the TID groups. If
	 * we cannot allocate the required number we exit and try again later
	 */
	if (kern_alloc_tids(flow))
		goto queue;
	/*
	 * Finally program the TID entries with the pagesets, compute the
	 * tidarray and enable the HW flow
	 */
	kern_program_rcvarray(flow);

	/*
	 * Setup the flow state with relevant information.
	 * This information is used for tracking the sequence of data packets
	 * for the segment.
	 * The flow is setup here as this is the most accurate time and place
	 * to do so. Doing at a later time runs the risk of the flow data in
	 * qpriv getting out of sync.
	 */
	memset(&flow->flow_state, 0x0, sizeof(flow->flow_state));
	flow->idx = qpriv->flow_state.index;
	flow->flow_state.generation = qpriv->flow_state.generation;
	flow->flow_state.spsn = qpriv->flow_state.psn;
	flow->flow_state.lpsn = flow->flow_state.spsn + flow->npkts - 1;
	flow->flow_state.r_next_psn =
		full_flow_psn(flow, flow->flow_state.spsn);
	qpriv->flow_state.psn += flow->npkts;

	dequeue_tid_waiter(rcd, &rcd->rarr_queue, flow->req->qp);
	/* get head before dropping lock */
	fqp = first_qp(rcd, &rcd->rarr_queue);
	spin_unlock_irqrestore(&rcd->exp_lock, flags);
	tid_rdma_schedule_tid_wakeup(fqp);

	req->setup_head = (req->setup_head + 1) & (MAX_FLOWS - 1);
	return 0;
queue:
	queue_qp_for_tid_wait(rcd, &rcd->rarr_queue, flow->req->qp);
	spin_unlock_irqrestore(&rcd->exp_lock, flags);
	return -EAGAIN;
}

static void hfi1_tid_rdma_reset_flow(struct tid_rdma_flow *flow)
{
	flow->npagesets = 0;
}

/*
 * This function is called after one segment has been successfully sent to
 * release the flow and TID HW/SW resources for that segment. The segments for a
 * TID RDMA request are setup and cleared in FIFO order which is managed using a
 * circular buffer.
 */
int hfi1_kern_exp_rcv_clear(struct tid_rdma_request *req)
	__must_hold(&req->qp->s_lock)
{
	struct tid_rdma_flow *flow = &req->flows[req->clear_tail];
	struct hfi1_ctxtdata *rcd = req->rcd;
	unsigned long flags;
	int i;
	struct rvt_qp *fqp;

	lockdep_assert_held(&req->qp->s_lock);
	/* Exit if we have nothing in the flow circular buffer */
	if (!CIRC_CNT(req->setup_head, req->clear_tail, MAX_FLOWS))
		return -EINVAL;

	spin_lock_irqsave(&rcd->exp_lock, flags);

	for (i = 0; i < flow->tnode_cnt; i++)
		kern_unprogram_rcv_group(flow, i);
	/* To prevent double unprogramming */
	flow->tnode_cnt = 0;
	/* get head before dropping lock */
	fqp = first_qp(rcd, &rcd->rarr_queue);
	spin_unlock_irqrestore(&rcd->exp_lock, flags);

	dma_unmap_flow(flow);

	hfi1_tid_rdma_reset_flow(flow);
	req->clear_tail = (req->clear_tail + 1) & (MAX_FLOWS - 1);

	if (fqp == req->qp) {
		__trigger_tid_waiter(fqp);
		rvt_put_qp(fqp);
	} else {
		tid_rdma_schedule_tid_wakeup(fqp);
	}

	return 0;
}

/*
 * This function is called to release all the tid entries for
 * a request.
 */
void hfi1_kern_exp_rcv_clear_all(struct tid_rdma_request *req)
	__must_hold(&req->qp->s_lock)
{
	/* Use memory barrier for proper ordering */
	while (CIRC_CNT(req->setup_head, req->clear_tail, MAX_FLOWS)) {
		if (hfi1_kern_exp_rcv_clear(req))
			break;
	}
}

/**
 * hfi1_kern_exp_rcv_free_flows - free priviously allocated flow information
 * @req - the tid rdma request to be cleaned
 */
static void hfi1_kern_exp_rcv_free_flows(struct tid_rdma_request *req)
{
	kfree(req->flows);
	req->flows = NULL;
}

/**
 * __trdma_clean_swqe - clean up for large sized QPs
 * @qp: the queue patch
 * @wqe: the send wqe
 */
void __trdma_clean_swqe(struct rvt_qp *qp, struct rvt_swqe *wqe)
{
	struct hfi1_swqe_priv *p = wqe->priv;

	hfi1_kern_exp_rcv_free_flows(&p->tid_req);
}

/*
 * This can be called at QP create time or in the data path.
 */
static int hfi1_kern_exp_rcv_alloc_flows(struct tid_rdma_request *req,
					 gfp_t gfp)
{
	struct tid_rdma_flow *flows;
	int i;

	if (likely(req->flows))
		return 0;
	flows = kmalloc_node(MAX_FLOWS * sizeof(*flows), gfp,
			     req->rcd->numa_id);
	if (!flows)
		return -ENOMEM;
	/* mini init */
	for (i = 0; i < MAX_FLOWS; i++) {
		flows[i].req = req;
		flows[i].npagesets = 0;
		flows[i].pagesets[0].mapped =  0;
		flows[i].resync_npkts = 0;
	}
	req->flows = flows;
	return 0;
}

static void hfi1_init_trdma_req(struct rvt_qp *qp,
				struct tid_rdma_request *req)
{
	struct hfi1_qp_priv *qpriv = qp->priv;

	/*
	 * Initialize various TID RDMA request variables.
	 * These variables are "static", which is why they
	 * can be pre-initialized here before the WRs has
	 * even been submitted.
	 * However, non-NULL values for these variables do not
	 * imply that this WQE has been enabled for TID RDMA.
	 * Drivers should check the WQE's opcode to determine
	 * if a request is a TID RDMA one or not.
	 */
	req->qp = qp;
	req->rcd = qpriv->rcd;
}

u64 hfi1_access_sw_tid_wait(const struct cntr_entry *entry,
			    void *context, int vl, int mode, u64 data)
{
	struct hfi1_devdata *dd = context;

	return dd->verbs_dev.n_tidwait;
}

static struct tid_rdma_flow *find_flow_ib(struct tid_rdma_request *req,
					  u32 psn, u16 *fidx)
{
	u16 head, tail;
	struct tid_rdma_flow *flow;

	head = req->setup_head;
	tail = req->clear_tail;
	for ( ; CIRC_CNT(head, tail, MAX_FLOWS);
	     tail = CIRC_NEXT(tail, MAX_FLOWS)) {
		flow = &req->flows[tail];
		if (cmp_psn(psn, flow->flow_state.ib_spsn) >= 0 &&
		    cmp_psn(psn, flow->flow_state.ib_lpsn) <= 0) {
			if (fidx)
				*fidx = tail;
			return flow;
		}
	}
	return NULL;
}

/* TID RDMA READ functions */
u32 hfi1_build_tid_rdma_read_packet(struct rvt_swqe *wqe,
				    struct ib_other_headers *ohdr, u32 *bth1,
				    u32 *bth2, u32 *len)
{
	struct tid_rdma_request *req = wqe_to_tid_req(wqe);
	struct tid_rdma_flow *flow = &req->flows[req->flow_idx];
	struct rvt_qp *qp = req->qp;
	struct hfi1_qp_priv *qpriv = qp->priv;
	struct hfi1_swqe_priv *wpriv = wqe->priv;
	struct tid_rdma_read_req *rreq = &ohdr->u.tid_rdma.r_req;
	struct tid_rdma_params *remote;
	u32 req_len = 0;
	void *req_addr = NULL;

	/* This is the IB psn used to send the request */
	*bth2 = mask_psn(flow->flow_state.ib_spsn + flow->pkt);
	trace_hfi1_tid_flow_build_read_pkt(qp, req->flow_idx, flow);

	/* TID Entries for TID RDMA READ payload */
	req_addr = &flow->tid_entry[flow->tid_idx];
	req_len = sizeof(*flow->tid_entry) *
			(flow->tidcnt - flow->tid_idx);

	memset(&ohdr->u.tid_rdma.r_req, 0, sizeof(ohdr->u.tid_rdma.r_req));
	wpriv->ss.sge.vaddr = req_addr;
	wpriv->ss.sge.sge_length = req_len;
	wpriv->ss.sge.length = wpriv->ss.sge.sge_length;
	/*
	 * We can safely zero these out. Since the first SGE covers the
	 * entire packet, nothing else should even look at the MR.
	 */
	wpriv->ss.sge.mr = NULL;
	wpriv->ss.sge.m = 0;
	wpriv->ss.sge.n = 0;

	wpriv->ss.sg_list = NULL;
	wpriv->ss.total_len = wpriv->ss.sge.sge_length;
	wpriv->ss.num_sge = 1;

	/* Construct the TID RDMA READ REQ packet header */
	rcu_read_lock();
	remote = rcu_dereference(qpriv->tid_rdma.remote);

	KDETH_RESET(rreq->kdeth0, KVER, 0x1);
	KDETH_RESET(rreq->kdeth1, JKEY, remote->jkey);
	rreq->reth.vaddr = cpu_to_be64(wqe->rdma_wr.remote_addr +
			   req->cur_seg * req->seg_len + flow->sent);
	rreq->reth.rkey = cpu_to_be32(wqe->rdma_wr.rkey);
	rreq->reth.length = cpu_to_be32(*len);
	rreq->tid_flow_psn =
		cpu_to_be32((flow->flow_state.generation <<
			     HFI1_KDETH_BTH_SEQ_SHIFT) |
			    ((flow->flow_state.spsn + flow->pkt) &
			     HFI1_KDETH_BTH_SEQ_MASK));
	rreq->tid_flow_qp =
		cpu_to_be32(qpriv->tid_rdma.local.qp |
			    ((flow->idx & TID_RDMA_DESTQP_FLOW_MASK) <<
			     TID_RDMA_DESTQP_FLOW_SHIFT) |
			    qpriv->rcd->ctxt);
	rreq->verbs_qp = cpu_to_be32(qp->remote_qpn);
	*bth1 &= ~RVT_QPN_MASK;
	*bth1 |= remote->qp;
	*bth2 |= IB_BTH_REQ_ACK;
	rcu_read_unlock();

	/* We are done with this segment */
	flow->sent += *len;
	req->cur_seg++;
	qp->s_state = TID_OP(READ_REQ);
	req->ack_pending++;
	req->flow_idx = (req->flow_idx + 1) & (MAX_FLOWS - 1);
	qpriv->pending_tid_r_segs++;
	qp->s_num_rd_atomic++;

	/* Set the TID RDMA READ request payload size */
	*len = req_len;

	return sizeof(ohdr->u.tid_rdma.r_req) / sizeof(u32);
}

/*
 * @len: contains the data length to read upon entry and the read request
 *       payload length upon exit.
 */
u32 hfi1_build_tid_rdma_read_req(struct rvt_qp *qp, struct rvt_swqe *wqe,
				 struct ib_other_headers *ohdr, u32 *bth1,
				 u32 *bth2, u32 *len)
	__must_hold(&qp->s_lock)
{
	struct hfi1_qp_priv *qpriv = qp->priv;
	struct tid_rdma_request *req = wqe_to_tid_req(wqe);
	struct tid_rdma_flow *flow = NULL;
	u32 hdwords = 0;
	bool last;
	bool retry = true;
	u32 npkts = rvt_div_round_up_mtu(qp, *len);

	trace_hfi1_tid_req_build_read_req(qp, 0, wqe->wr.opcode, wqe->psn,
					  wqe->lpsn, req);
	/*
	 * Check sync conditions. Make sure that there are no pending
	 * segments before freeing the flow.
	 */
sync_check:
	if (req->state == TID_REQUEST_SYNC) {
		if (qpriv->pending_tid_r_segs)
			goto done;

		hfi1_kern_clear_hw_flow(req->rcd, qp);
		qpriv->s_flags &= ~HFI1_R_TID_SW_PSN;
		req->state = TID_REQUEST_ACTIVE;
	}

	/*
	 * If the request for this segment is resent, the tid resources should
	 * have been allocated before. In this case, req->flow_idx should
	 * fall behind req->setup_head.
	 */
	if (req->flow_idx == req->setup_head) {
		retry = false;
		if (req->state == TID_REQUEST_RESEND) {
			/*
			 * This is the first new segment for a request whose
			 * earlier segments have been re-sent. We need to
			 * set up the sge pointer correctly.
			 */
			restart_sge(&qp->s_sge, wqe, req->s_next_psn,
				    qp->pmtu);
			req->isge = 0;
			req->state = TID_REQUEST_ACTIVE;
		}

		/*
		 * Check sync. The last PSN of each generation is reserved for
		 * RESYNC.
		 */
		if ((qpriv->flow_state.psn + npkts) > MAX_TID_FLOW_PSN - 1) {
			req->state = TID_REQUEST_SYNC;
			goto sync_check;
		}

		/* Allocate the flow if not yet */
		if (hfi1_kern_setup_hw_flow(qpriv->rcd, qp))
			goto done;

		/*
		 * The following call will advance req->setup_head after
		 * allocating the tid entries.
		 */
		if (hfi1_kern_exp_rcv_setup(req, &qp->s_sge, &last)) {
			req->state = TID_REQUEST_QUEUED;

			/*
			 * We don't have resources for this segment. The QP has
			 * already been queued.
			 */
			goto done;
		}
	}

	/* req->flow_idx should only be one slot behind req->setup_head */
	flow = &req->flows[req->flow_idx];
	flow->pkt = 0;
	flow->tid_idx = 0;
	flow->sent = 0;
	if (!retry) {
		/* Set the first and last IB PSN for the flow in use.*/
		flow->flow_state.ib_spsn = req->s_next_psn;
		flow->flow_state.ib_lpsn =
			flow->flow_state.ib_spsn + flow->npkts - 1;
	}

	/* Calculate the next segment start psn.*/
	req->s_next_psn += flow->npkts;

	/* Build the packet header */
	hdwords = hfi1_build_tid_rdma_read_packet(wqe, ohdr, bth1, bth2, len);
done:
	return hdwords;
}

/*
 * Validate and accept the TID RDMA READ request parameters.
 * Return 0 if the request is accepted successfully;
 * Return 1 otherwise.
 */
static int tid_rdma_rcv_read_request(struct rvt_qp *qp,
				     struct rvt_ack_entry *e,
				     struct hfi1_packet *packet,
				     struct ib_other_headers *ohdr,
				     u32 bth0, u32 psn, u64 vaddr, u32 len)
{
	struct hfi1_qp_priv *qpriv = qp->priv;
	struct tid_rdma_request *req;
	struct tid_rdma_flow *flow;
	u32 flow_psn, i, tidlen = 0, pktlen, tlen;

	req = ack_to_tid_req(e);

	/* Validate the payload first */
	flow = &req->flows[req->setup_head];

	/* payload length = packet length - (header length + ICRC length) */
	pktlen = packet->tlen - (packet->hlen + 4);
	if (pktlen > sizeof(flow->tid_entry))
		return 1;
	memcpy(flow->tid_entry, packet->ebuf, pktlen);
	flow->tidcnt = pktlen / sizeof(*flow->tid_entry);

	/*
	 * Walk the TID_ENTRY list to make sure we have enough space for a
	 * complete segment. Also calculate the number of required packets.
	 */
	flow->npkts = rvt_div_round_up_mtu(qp, len);
	for (i = 0; i < flow->tidcnt; i++) {
		trace_hfi1_tid_entry_rcv_read_req(qp, i,
						  flow->tid_entry[i]);
		tlen = EXP_TID_GET(flow->tid_entry[i], LEN);
		if (!tlen)
			return 1;

		/*
		 * For tid pair (tidctr == 3), the buffer size of the pair
		 * should be the sum of the buffer size described by each
		 * tid entry. However, only the first entry needs to be
		 * specified in the request (see WFR HAS Section 8.5.7.1).
		 */
		tidlen += tlen;
	}
	if (tidlen * PAGE_SIZE < len)
		return 1;

	/* Empty the flow array */
	req->clear_tail = req->setup_head;
	flow->pkt = 0;
	flow->tid_idx = 0;
	flow->tid_offset = 0;
	flow->sent = 0;
	flow->tid_qpn = be32_to_cpu(ohdr->u.tid_rdma.r_req.tid_flow_qp);
	flow->idx = (flow->tid_qpn >> TID_RDMA_DESTQP_FLOW_SHIFT) &
		    TID_RDMA_DESTQP_FLOW_MASK;
	flow_psn = mask_psn(be32_to_cpu(ohdr->u.tid_rdma.r_req.tid_flow_psn));
	flow->flow_state.generation = flow_psn >> HFI1_KDETH_BTH_SEQ_SHIFT;
	flow->flow_state.spsn = flow_psn & HFI1_KDETH_BTH_SEQ_MASK;
	flow->length = len;

	flow->flow_state.lpsn = flow->flow_state.spsn +
		flow->npkts - 1;
	flow->flow_state.ib_spsn = psn;
	flow->flow_state.ib_lpsn = flow->flow_state.ib_spsn + flow->npkts - 1;

	trace_hfi1_tid_flow_rcv_read_req(qp, req->setup_head, flow);
	/* Set the initial flow index to the current flow. */
	req->flow_idx = req->setup_head;

	/* advance circular buffer head */
	req->setup_head = (req->setup_head + 1) & (MAX_FLOWS - 1);

	/*
	 * Compute last PSN for request.
	 */
	e->opcode = (bth0 >> 24) & 0xff;
	e->psn = psn;
	e->lpsn = psn + flow->npkts - 1;
	e->sent = 0;

	req->n_flows = qpriv->tid_rdma.local.max_read;
	req->state = TID_REQUEST_ACTIVE;
	req->cur_seg = 0;
	req->comp_seg = 0;
	req->ack_seg = 0;
	req->isge = 0;
	req->seg_len = qpriv->tid_rdma.local.max_len;
	req->total_len = len;
	req->total_segs = 1;
	req->r_flow_psn = e->psn;

	trace_hfi1_tid_req_rcv_read_req(qp, 0, e->opcode, e->psn, e->lpsn,
					req);
	return 0;
}

static int tid_rdma_rcv_error(struct hfi1_packet *packet,
			      struct ib_other_headers *ohdr,
			      struct rvt_qp *qp, u32 psn, int diff)
{
	struct hfi1_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
	struct hfi1_ctxtdata *rcd = ((struct hfi1_qp_priv *)qp->priv)->rcd;
	struct hfi1_ibdev *dev = to_idev(qp->ibqp.device);
	struct hfi1_qp_priv *qpriv = qp->priv;
	struct rvt_ack_entry *e;
	struct tid_rdma_request *req;
	unsigned long flags;
	u8 prev;
	bool old_req;

	trace_hfi1_rsp_tid_rcv_error(qp, psn);
	trace_hfi1_tid_rdma_rcv_err(qp, 0, psn, diff);
	if (diff > 0) {
		/* sequence error */
		if (!qp->r_nak_state) {
			ibp->rvp.n_rc_seqnak++;
			qp->r_nak_state = IB_NAK_PSN_ERROR;
			qp->r_ack_psn = qp->r_psn;
			rc_defered_ack(rcd, qp);
		}
		goto done;
	}

	ibp->rvp.n_rc_dupreq++;

	spin_lock_irqsave(&qp->s_lock, flags);
	e = find_prev_entry(qp, psn, &prev, NULL, &old_req);
	if (!e || (e->opcode != TID_OP(READ_REQ) &&
		   e->opcode != TID_OP(WRITE_REQ)))
		goto unlock;

	req = ack_to_tid_req(e);
	req->r_flow_psn = psn;
	trace_hfi1_tid_req_rcv_err(qp, 0, e->opcode, e->psn, e->lpsn, req);
	if (e->opcode == TID_OP(READ_REQ)) {
		struct ib_reth *reth;
		u32 offset;
		u32 len;
		u32 rkey;
		u64 vaddr;
		int ok;
		u32 bth0;

		reth = &ohdr->u.tid_rdma.r_req.reth;
		/*
		 * The requester always restarts from the start of the original
		 * request.
		 */
		offset = delta_psn(psn, e->psn) * qp->pmtu;
		len = be32_to_cpu(reth->length);
		if (psn != e->psn || len != req->total_len)
			goto unlock;

		release_rdma_sge_mr(e);

		rkey = be32_to_cpu(reth->rkey);
		vaddr = get_ib_reth_vaddr(reth);

		qp->r_len = len;
		ok = rvt_rkey_ok(qp, &e->rdma_sge, len, vaddr, rkey,
				 IB_ACCESS_REMOTE_READ);
		if (unlikely(!ok))
			goto unlock;

		/*
		 * If all the response packets for the current request have
		 * been sent out and this request is complete (old_request
		 * == false) and the TID flow may be unusable (the
		 * req->clear_tail is advanced). However, when an earlier
		 * request is received, this request will not be complete any
		 * more (qp->s_tail_ack_queue is moved back, see below).
		 * Consequently, we need to update the TID flow info everytime
		 * a duplicate request is received.
		 */
		bth0 = be32_to_cpu(ohdr->bth[0]);
		if (tid_rdma_rcv_read_request(qp, e, packet, ohdr, bth0, psn,
					      vaddr, len))
			goto unlock;

		/*
		 * True if the request is already scheduled (between
		 * qp->s_tail_ack_queue and qp->r_head_ack_queue);
		 */
		if (old_req)
			goto unlock;
	} else {
		struct flow_state *fstate;
		bool schedule = false;
		u8 i;

		if (req->state == TID_REQUEST_RESEND) {
			req->state = TID_REQUEST_RESEND_ACTIVE;
		} else if (req->state == TID_REQUEST_INIT_RESEND) {
			req->state = TID_REQUEST_INIT;
			schedule = true;
		}

		/*
		 * True if the request is already scheduled (between
		 * qp->s_tail_ack_queue and qp->r_head_ack_queue).
		 * Also, don't change requests, which are at the SYNC
		 * point and haven't generated any responses yet.
		 * There is nothing to retransmit for them yet.
		 */
		if (old_req || req->state == TID_REQUEST_INIT ||
		    (req->state == TID_REQUEST_SYNC && !req->cur_seg)) {
			for (i = prev + 1; ; i++) {
				if (i > rvt_size_atomic(&dev->rdi))
					i = 0;
				if (i == qp->r_head_ack_queue)
					break;
				e = &qp->s_ack_queue[i];
				req = ack_to_tid_req(e);
				if (e->opcode == TID_OP(WRITE_REQ) &&
				    req->state == TID_REQUEST_INIT)
					req->state = TID_REQUEST_INIT_RESEND;
			}
			/*
			 * If the state of the request has been changed,
			 * the first leg needs to get scheduled in order to
			 * pick up the change. Otherwise, normal response
			 * processing should take care of it.
			 */
			if (!schedule)
				goto unlock;
		}

		/*
		 * If there is no more allocated segment, just schedule the qp
		 * without changing any state.
		 */
		if (req->clear_tail == req->setup_head)
			goto schedule;
		/*
		 * If this request has sent responses for segments, which have
		 * not received data yet (flow_idx != clear_tail), the flow_idx
		 * pointer needs to be adjusted so the same responses can be
		 * re-sent.
		 */
		if (CIRC_CNT(req->flow_idx, req->clear_tail, MAX_FLOWS)) {
			fstate = &req->flows[req->clear_tail].flow_state;
			qpriv->pending_tid_w_segs -=
				CIRC_CNT(req->flow_idx, req->clear_tail,
					 MAX_FLOWS);
			req->flow_idx =
				CIRC_ADD(req->clear_tail,
					 delta_psn(psn, fstate->resp_ib_psn),
					 MAX_FLOWS);
			qpriv->pending_tid_w_segs +=
				delta_psn(psn, fstate->resp_ib_psn);
			/*
			 * When flow_idx == setup_head, we've gotten a duplicate
			 * request for a segment, which has not been allocated
			 * yet. In that case, don't adjust this request.
			 * However, we still want to go through the loop below
			 * to adjust all subsequent requests.
			 */
			if (CIRC_CNT(req->setup_head, req->flow_idx,
				     MAX_FLOWS)) {
				req->cur_seg = delta_psn(psn, e->psn);
				req->state = TID_REQUEST_RESEND_ACTIVE;
			}
		}

		for (i = prev + 1; ; i++) {
			/*
			 * Look at everything up to and including
			 * s_tail_ack_queue
			 */
			if (i > rvt_size_atomic(&dev->rdi))
				i = 0;
			if (i == qp->r_head_ack_queue)
				break;
			e = &qp->s_ack_queue[i];
			req = ack_to_tid_req(e);
			trace_hfi1_tid_req_rcv_err(qp, 0, e->opcode, e->psn,
						   e->lpsn, req);
			if (e->opcode != TID_OP(WRITE_REQ) ||
			    req->cur_seg == req->comp_seg ||
			    req->state == TID_REQUEST_INIT ||
			    req->state == TID_REQUEST_INIT_RESEND) {
				if (req->state == TID_REQUEST_INIT)
					req->state = TID_REQUEST_INIT_RESEND;
				continue;
			}
			qpriv->pending_tid_w_segs -=
				CIRC_CNT(req->flow_idx,
					 req->clear_tail,
					 MAX_FLOWS);
			req->flow_idx = req->clear_tail;
			req->state = TID_REQUEST_RESEND;
			req->cur_seg = req->comp_seg;
		}
		qpriv->s_flags &= ~HFI1_R_TID_WAIT_INTERLCK;
	}
	/* Re-process old requests.*/
	if (qp->s_acked_ack_queue == qp->s_tail_ack_queue)
		qp->s_acked_ack_queue = prev;
	qp->s_tail_ack_queue = prev;
	/*
	 * Since the qp->s_tail_ack_queue is modified, the
	 * qp->s_ack_state must be changed to re-initialize
	 * qp->s_ack_rdma_sge; Otherwise, we will end up in
	 * wrong memory region.
	 */
	qp->s_ack_state = OP(ACKNOWLEDGE);
schedule:
	/*
	 * It's possible to receive a retry psn that is earlier than an RNRNAK
	 * psn. In this case, the rnrnak state should be cleared.
	 */
	if (qpriv->rnr_nak_state) {
		qp->s_nak_state = 0;
		qpriv->rnr_nak_state = TID_RNR_NAK_INIT;
		qp->r_psn = e->lpsn + 1;
		hfi1_tid_write_alloc_resources(qp, true);
	}

	qp->r_state = e->opcode;
	qp->r_nak_state = 0;
	qp->s_flags |= RVT_S_RESP_PENDING;
	hfi1_schedule_send(qp);
unlock:
	spin_unlock_irqrestore(&qp->s_lock, flags);
done:
	return 1;
}

void hfi1_rc_rcv_tid_rdma_read_req(struct hfi1_packet *packet)
{
	/* HANDLER FOR TID RDMA READ REQUEST packet (Responder side)*/

	/*
	 * 1. Verify TID RDMA READ REQ as per IB_OPCODE_RC_RDMA_READ
	 *    (see hfi1_rc_rcv())
	 * 2. Put TID RDMA READ REQ into the response queueu (s_ack_queue)
	 *     - Setup struct tid_rdma_req with request info
	 *     - Initialize struct tid_rdma_flow info;
	 *     - Copy TID entries;
	 * 3. Set the qp->s_ack_state.
	 * 4. Set RVT_S_RESP_PENDING in s_flags.
	 * 5. Kick the send engine (hfi1_schedule_send())
	 */
	struct hfi1_ctxtdata *rcd = packet->rcd;
	struct rvt_qp *qp = packet->qp;
	struct hfi1_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
	struct ib_other_headers *ohdr = packet->ohdr;
	struct rvt_ack_entry *e;
	unsigned long flags;
	struct ib_reth *reth;
	struct hfi1_qp_priv *qpriv = qp->priv;
	u32 bth0, psn, len, rkey;
	bool fecn;
	u8 next;
	u64 vaddr;
	int diff;
	u8 nack_state = IB_NAK_INVALID_REQUEST;

	bth0 = be32_to_cpu(ohdr->bth[0]);
	if (hfi1_ruc_check_hdr(ibp, packet))
		return;

	fecn = process_ecn(qp, packet);
	psn = mask_psn(be32_to_cpu(ohdr->bth[2]));
	trace_hfi1_rsp_rcv_tid_read_req(qp, psn);

	if (qp->state == IB_QPS_RTR && !(qp->r_flags & RVT_R_COMM_EST))
		rvt_comm_est(qp);

	if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_READ)))
		goto nack_inv;

	reth = &ohdr->u.tid_rdma.r_req.reth;
	vaddr = be64_to_cpu(reth->vaddr);
	len = be32_to_cpu(reth->length);
	/* The length needs to be in multiples of PAGE_SIZE */
	if (!len || len & ~PAGE_MASK || len > qpriv->tid_rdma.local.max_len)
		goto nack_inv;

	diff = delta_psn(psn, qp->r_psn);
	if (unlikely(diff)) {
		tid_rdma_rcv_err(packet, ohdr, qp, psn, diff, fecn);
		return;
	}

	/* We've verified the request, insert it into the ack queue. */
	next = qp->r_head_ack_queue + 1;
	if (next > rvt_size_atomic(ib_to_rvt(qp->ibqp.device)))
		next = 0;
	spin_lock_irqsave(&qp->s_lock, flags);
	if (unlikely(next == qp->s_tail_ack_queue)) {
		if (!qp->s_ack_queue[next].sent) {
			nack_state = IB_NAK_REMOTE_OPERATIONAL_ERROR;
			goto nack_inv_unlock;
		}
		update_ack_queue(qp, next);
	}
	e = &qp->s_ack_queue[qp->r_head_ack_queue];
	release_rdma_sge_mr(e);

	rkey = be32_to_cpu(reth->rkey);
	qp->r_len = len;

	if (unlikely(!rvt_rkey_ok(qp, &e->rdma_sge, qp->r_len, vaddr,
				  rkey, IB_ACCESS_REMOTE_READ)))
		goto nack_acc;

	/* Accept the request parameters */
	if (tid_rdma_rcv_read_request(qp, e, packet, ohdr, bth0, psn, vaddr,
				      len))
		goto nack_inv_unlock;

	qp->r_state = e->opcode;
	qp->r_nak_state = 0;
	/*
	 * We need to increment the MSN here instead of when we
	 * finish sending the result since a duplicate request would
	 * increment it more than once.
	 */
	qp->r_msn++;
	qp->r_psn += e->lpsn - e->psn + 1;

	qp->r_head_ack_queue = next;

	/*
	 * For all requests other than TID WRITE which are added to the ack
	 * queue, qpriv->r_tid_alloc follows qp->r_head_ack_queue. It is ok to
	 * do this because of interlocks between these and TID WRITE
	 * requests. The same change has also been made in hfi1_rc_rcv().
	 */
	qpriv->r_tid_alloc = qp->r_head_ack_queue;

	/* Schedule the send tasklet. */
	qp->s_flags |= RVT_S_RESP_PENDING;
	if (fecn)
		qp->s_flags |= RVT_S_ECN;
	hfi1_schedule_send(qp);

	spin_unlock_irqrestore(&qp->s_lock, flags);
	return;

nack_inv_unlock:
	spin_unlock_irqrestore(&qp->s_lock, flags);
nack_inv:
	rvt_rc_error(qp, IB_WC_LOC_QP_OP_ERR);
	qp->r_nak_state = nack_state;
	qp->r_ack_psn = qp->r_psn;
	/* Queue NAK for later */
	rc_defered_ack(rcd, qp);
	return;
nack_acc:
	spin_unlock_irqrestore(&qp->s_lock, flags);
	rvt_rc_error(qp, IB_WC_LOC_PROT_ERR);
	qp->r_nak_state = IB_NAK_REMOTE_ACCESS_ERROR;
	qp->r_ack_psn = qp->r_psn;
}

u32 hfi1_build_tid_rdma_read_resp(struct rvt_qp *qp, struct rvt_ack_entry *e,
				  struct ib_other_headers *ohdr, u32 *bth0,
				  u32 *bth1, u32 *bth2, u32 *len, bool *last)
{
	struct hfi1_ack_priv *epriv = e->priv;
	struct tid_rdma_request *req = &epriv->tid_req;
	struct hfi1_qp_priv *qpriv = qp->priv;
	struct tid_rdma_flow *flow = &req->flows[req->clear_tail];
	u32 tidentry = flow->tid_entry[flow->tid_idx];
	u32 tidlen = EXP_TID_GET(tidentry, LEN) << PAGE_SHIFT;
	struct tid_rdma_read_resp *resp = &ohdr->u.tid_rdma.r_rsp;
	u32 next_offset, om = KDETH_OM_LARGE;
	bool last_pkt;
	u32 hdwords = 0;
	struct tid_rdma_params *remote;

	*len = min_t(u32, qp->pmtu, tidlen - flow->tid_offset);
	flow->sent += *len;
	next_offset = flow->tid_offset + *len;
	last_pkt = (flow->sent >= flow->length);

	trace_hfi1_tid_entry_build_read_resp(qp, flow->tid_idx, tidentry);
	trace_hfi1_tid_flow_build_read_resp(qp, req->clear_tail, flow);

	rcu_read_lock();
	remote = rcu_dereference(qpriv->tid_rdma.remote);
	if (!remote) {
		rcu_read_unlock();
		goto done;
	}
	KDETH_RESET(resp->kdeth0, KVER, 0x1);
	KDETH_SET(resp->kdeth0, SH, !last_pkt);
	KDETH_SET(resp->kdeth0, INTR, !!(!last_pkt && remote->urg));
	KDETH_SET(resp->kdeth0, TIDCTRL, EXP_TID_GET(tidentry, CTRL));
	KDETH_SET(resp->kdeth0, TID, EXP_TID_GET(tidentry, IDX));
	KDETH_SET(resp->kdeth0, OM, om == KDETH_OM_LARGE);
	KDETH_SET(resp->kdeth0, OFFSET, flow->tid_offset / om);
	KDETH_RESET(resp->kdeth1, JKEY, remote->jkey);
	resp->verbs_qp = cpu_to_be32(qp->remote_qpn);
	rcu_read_unlock();

	resp->aeth = rvt_compute_aeth(qp);
	resp->verbs_psn = cpu_to_be32(mask_psn(flow->flow_state.ib_spsn +
					       flow->pkt));

	*bth0 = TID_OP(READ_RESP) << 24;
	*bth1 = flow->tid_qpn;
	*bth2 = mask_psn(((flow->flow_state.spsn + flow->pkt++) &
			  HFI1_KDETH_BTH_SEQ_MASK) |
			 (flow->flow_state.generation <<
			  HFI1_KDETH_BTH_SEQ_SHIFT));
	*last = last_pkt;
	if (last_pkt)
		/* Advance to next flow */
		req->clear_tail = (req->clear_tail + 1) &
				  (MAX_FLOWS - 1);

	if (next_offset >= tidlen) {
		flow->tid_offset = 0;
		flow->tid_idx++;
	} else {
		flow->tid_offset = next_offset;
	}

	hdwords = sizeof(ohdr->u.tid_rdma.r_rsp) / sizeof(u32);

done:
	return hdwords;
}

static inline struct tid_rdma_request *
find_tid_request(struct rvt_qp *qp, u32 psn, enum ib_wr_opcode opcode)
	__must_hold(&qp->s_lock)
{
	struct rvt_swqe *wqe;
	struct tid_rdma_request *req = NULL;
	u32 i, end;

	end = qp->s_cur + 1;
	if (end == qp->s_size)
		end = 0;
	for (i = qp->s_acked; i != end;) {
		wqe = rvt_get_swqe_ptr(qp, i);
		if (cmp_psn(psn, wqe->psn) >= 0 &&
		    cmp_psn(psn, wqe->lpsn) <= 0) {
			if (wqe->wr.opcode == opcode)
				req = wqe_to_tid_req(wqe);
			break;
		}
		if (++i == qp->s_size)
			i = 0;
	}

	return req;
}

void hfi1_rc_rcv_tid_rdma_read_resp(struct hfi1_packet *packet)
{
	/* HANDLER FOR TID RDMA READ RESPONSE packet (Requestor side */

	/*
	 * 1. Find matching SWQE
	 * 2. Check that the entire segment has been read.
	 * 3. Remove HFI1_S_WAIT_TID_RESP from s_flags.
	 * 4. Free the TID flow resources.
	 * 5. Kick the send engine (hfi1_schedule_send())
	 */
	struct ib_other_headers *ohdr = packet->ohdr;
	struct rvt_qp *qp = packet->qp;
	struct hfi1_qp_priv *priv = qp->priv;
	struct hfi1_ctxtdata *rcd = packet->rcd;
	struct tid_rdma_request *req;
	struct tid_rdma_flow *flow;
	u32 opcode, aeth;
	bool fecn;
	unsigned long flags;
	u32 kpsn, ipsn;

	trace_hfi1_sender_rcv_tid_read_resp(qp);
	fecn = process_ecn(qp, packet);
	kpsn = mask_psn(be32_to_cpu(ohdr->bth[2]));
	aeth = be32_to_cpu(ohdr->u.tid_rdma.r_rsp.aeth);
	opcode = (be32_to_cpu(ohdr->bth[0]) >> 24) & 0xff;

	spin_lock_irqsave(&qp->s_lock, flags);
	ipsn = mask_psn(be32_to_cpu(ohdr->u.tid_rdma.r_rsp.verbs_psn));
	req = find_tid_request(qp, ipsn, IB_WR_TID_RDMA_READ);
	if (unlikely(!req))
		goto ack_op_err;

	flow = &req->flows[req->clear_tail];
	/* When header suppression is disabled */
	if (cmp_psn(ipsn, flow->flow_state.ib_lpsn)) {
		update_r_next_psn_fecn(packet, priv, rcd, flow, fecn);

		if (cmp_psn(kpsn, flow->flow_state.r_next_psn))
			goto ack_done;
		flow->flow_state.r_next_psn = mask_psn(kpsn + 1);
		/*
		 * Copy the payload to destination buffer if this packet is
		 * delivered as an eager packet due to RSM rule and FECN.
		 * The RSM rule selects FECN bit in BTH and SH bit in
		 * KDETH header and therefore will not match the last
		 * packet of each segment that has SH bit cleared.
		 */
		if (fecn && packet->etype == RHF_RCV_TYPE_EAGER) {
			struct rvt_sge_state ss;
			u32 len;
			u32 tlen = packet->tlen;
			u16 hdrsize = packet->hlen;
			u8 pad = packet->pad;
			u8 extra_bytes = pad + packet->extra_byte +
				(SIZE_OF_CRC << 2);
			u32 pmtu = qp->pmtu;

			if (unlikely(tlen != (hdrsize + pmtu + extra_bytes)))
				goto ack_op_err;
			len = restart_sge(&ss, req->e.swqe, ipsn, pmtu);
			if (unlikely(len < pmtu))
				goto ack_op_err;
			rvt_copy_sge(qp, &ss, packet->payload, pmtu, false,
				     false);
			/* Raise the sw sequence check flag for next packet */
			priv->s_flags |= HFI1_R_TID_SW_PSN;
		}

		goto ack_done;
	}
	flow->flow_state.r_next_psn = mask_psn(kpsn + 1);
	req->ack_pending--;
	priv->pending_tid_r_segs--;
	qp->s_num_rd_atomic--;
	if ((qp->s_flags & RVT_S_WAIT_FENCE) &&
	    !qp->s_num_rd_atomic) {
		qp->s_flags &= ~(RVT_S_WAIT_FENCE |
				 RVT_S_WAIT_ACK);
		hfi1_schedule_send(qp);
	}
	if (qp->s_flags & RVT_S_WAIT_RDMAR) {
		qp->s_flags &= ~(RVT_S_WAIT_RDMAR | RVT_S_WAIT_ACK);
		hfi1_schedule_send(qp);
	}

	trace_hfi1_ack(qp, ipsn);
	trace_hfi1_tid_req_rcv_read_resp(qp, 0, req->e.swqe->wr.opcode,
					 req->e.swqe->psn, req->e.swqe->lpsn,
					 req);
	trace_hfi1_tid_flow_rcv_read_resp(qp, req->clear_tail, flow);

	/* Release the tid resources */
	hfi1_kern_exp_rcv_clear(req);

	if (!do_rc_ack(qp, aeth, ipsn, opcode, 0, rcd))
		goto ack_done;

	/* If not done yet, build next read request */
	if (++req->comp_seg >= req->total_segs) {
		priv->tid_r_comp++;
		req->state = TID_REQUEST_COMPLETE;
	}

	/*
	 * Clear the hw flow under two conditions:
	 * 1. This request is a sync point and it is complete;
	 * 2. Current request is completed and there are no more requests.
	 */
	if ((req->state == TID_REQUEST_SYNC &&
	     req->comp_seg == req->cur_seg) ||
	    priv->tid_r_comp == priv->tid_r_reqs) {
		hfi1_kern_clear_hw_flow(priv->rcd, qp);
		priv->s_flags &= ~HFI1_R_TID_SW_PSN;
		if (req->state == TID_REQUEST_SYNC)
			req->state = TID_REQUEST_ACTIVE;
	}

	hfi1_schedule_send(qp);
	goto ack_done;

ack_op_err:
	/*
	 * The test indicates that the send engine has finished its cleanup
	 * after sending the request and it's now safe to put the QP into error
	 * state. However, if the wqe queue is empty (qp->s_acked == qp->s_tail
	 * == qp->s_head), it would be unsafe to complete the wqe pointed by
	 * qp->s_acked here. Putting the qp into error state will safely flush
	 * all remaining requests.
	 */
	if (qp->s_last == qp->s_acked)
		rvt_error_qp(qp, IB_WC_WR_FLUSH_ERR);

ack_done:
	spin_unlock_irqrestore(&qp->s_lock, flags);
}

void hfi1_kern_read_tid_flow_free(struct rvt_qp *qp)
	__must_hold(&qp->s_lock)
{
	u32 n = qp->s_acked;
	struct rvt_swqe *wqe;
	struct tid_rdma_request *req;
	struct hfi1_qp_priv *priv = qp->priv;

	lockdep_assert_held(&qp->s_lock);
	/* Free any TID entries */
	while (n != qp->s_tail) {
		wqe = rvt_get_swqe_ptr(qp, n);
		if (wqe->wr.opcode == IB_WR_TID_RDMA_READ) {
			req = wqe_to_tid_req(wqe);
			hfi1_kern_exp_rcv_clear_all(req);
		}

		if (++n == qp->s_size)
			n = 0;
	}
	/* Free flow */
	hfi1_kern_clear_hw_flow(priv->rcd, qp);
}

static bool tid_rdma_tid_err(struct hfi1_packet *packet, u8 rcv_type)
{
	struct rvt_qp *qp = packet->qp;

	if (rcv_type >= RHF_RCV_TYPE_IB)
		goto done;

	spin_lock(&qp->s_lock);

	/*
	 * We've ran out of space in the eager buffer.
	 * Eagerly received KDETH packets which require space in the
	 * Eager buffer (packet that have payload) are TID RDMA WRITE
	 * response packets. In this case, we have to re-transmit the
	 * TID RDMA WRITE request.
	 */
	if (rcv_type == RHF_RCV_TYPE_EAGER) {
		hfi1_restart_rc(qp, qp->s_last_psn + 1, 1);
		hfi1_schedule_send(qp);
	}

	/* Since no payload is delivered, just drop the packet */
	spin_unlock(&qp->s_lock);
done:
	return true;
}

static void restart_tid_rdma_read_req(struct hfi1_ctxtdata *rcd,
				      struct rvt_qp *qp, struct rvt_swqe *wqe)
{
	struct tid_rdma_request *req;
	struct tid_rdma_flow *flow;

	/* Start from the right segment */
	qp->r_flags |= RVT_R_RDMAR_SEQ;
	req = wqe_to_tid_req(wqe);
	flow = &req->flows[req->clear_tail];
	hfi1_restart_rc(qp, flow->flow_state.ib_spsn, 0);
	if (list_empty(&qp->rspwait)) {
		qp->r_flags |= RVT_R_RSP_SEND;
		rvt_get_qp(qp);
		list_add_tail(&qp->rspwait, &rcd->qp_wait_list);
	}
}

/*
 * Handle the KDETH eflags for TID RDMA READ response.
 *
 * Return true if the last packet for a segment has been received and it is
 * time to process the response normally; otherwise, return true.
 *
 * The caller must hold the packet->qp->r_lock and the rcu_read_lock.
 */
static bool handle_read_kdeth_eflags(struct hfi1_ctxtdata *rcd,
				     struct hfi1_packet *packet, u8 rcv_type,
				     u8 rte, u32 psn, u32 ibpsn)
	__must_hold(&packet->qp->r_lock) __must_hold(RCU)
{
	struct hfi1_pportdata *ppd = rcd->ppd;
	struct hfi1_devdata *dd = ppd->dd;
	struct hfi1_ibport *ibp;
	struct rvt_swqe *wqe;
	struct tid_rdma_request *req;
	struct tid_rdma_flow *flow;
	u32 ack_psn;
	struct rvt_qp *qp = packet->qp;
	struct hfi1_qp_priv *priv = qp->priv;
	bool ret = true;
	int diff = 0;
	u32 fpsn;

	lockdep_assert_held(&qp->r_lock);
	spin_lock(&qp->s_lock);
	/* If the psn is out of valid range, drop the packet */
	if (cmp_psn(ibpsn, qp->s_last_psn) < 0 ||
	    cmp_psn(ibpsn, qp->s_psn) > 0)
		goto s_unlock;

	/*
	 * Note that NAKs implicitly ACK outstanding SEND and RDMA write
	 * requests and implicitly NAK RDMA read and atomic requests issued
	 * before the NAK'ed request.
	 */
	ack_psn = ibpsn - 1;
	wqe = rvt_get_swqe_ptr(qp, qp->s_acked);
	ibp = to_iport(qp->ibqp.device, qp->port_num);

	/* Complete WQEs that the PSN finishes. */
	while ((int)delta_psn(ack_psn, wqe->lpsn) >= 0) {
		/*
		 * If this request is a RDMA read or atomic, and the NACK is
		 * for a later operation, this NACK NAKs the RDMA read or
		 * atomic.
		 */
		if (wqe->wr.opcode == IB_WR_RDMA_READ ||
		    wqe->wr.opcode == IB_WR_TID_RDMA_READ ||
		    wqe->wr.opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
		    wqe->wr.opcode == IB_WR_ATOMIC_FETCH_AND_ADD) {
			/* Retry this request. */
			if (!(qp->r_flags & RVT_R_RDMAR_SEQ)) {
				qp->r_flags |= RVT_R_RDMAR_SEQ;
				if (wqe->wr.opcode == IB_WR_TID_RDMA_READ) {
					restart_tid_rdma_read_req(rcd, qp,
								  wqe);
				} else {
					hfi1_restart_rc(qp, qp->s_last_psn + 1,
							0);
					if (list_empty(&qp->rspwait)) {
						qp->r_flags |= RVT_R_RSP_SEND;
						rvt_get_qp(qp);
						list_add_tail(/* wait */
						   &qp->rspwait,
						   &rcd->qp_wait_list);
					}
				}
			}
			/*
			 * No need to process the NAK since we are
			 * restarting an earlier request.
			 */
			break;
		}

		wqe = do_rc_completion(qp, wqe, ibp);
		if (qp->s_acked == qp->s_tail)
			goto s_unlock;
	}

	if (qp->s_acked == qp->s_tail)
		goto s_unlock;

	/* Handle the eflags for the request */
	if (wqe->wr.opcode != IB_WR_TID_RDMA_READ)
		goto s_unlock;

	req = wqe_to_tid_req(wqe);
	switch (rcv_type) {
	case RHF_RCV_TYPE_EXPECTED:
		switch (rte) {
		case RHF_RTE_EXPECTED_FLOW_SEQ_ERR:
			/*
			 * On the first occurrence of a Flow Sequence error,
			 * the flag TID_FLOW_SW_PSN is set.
			 *
			 * After that, the flow is *not* reprogrammed and the
			 * protocol falls back to SW PSN checking. This is done
			 * to prevent continuous Flow Sequence errors for any
			 * packets that could be still in the fabric.
			 */
			flow = &req->flows[req->clear_tail];
			if (priv->s_flags & HFI1_R_TID_SW_PSN) {
				diff = cmp_psn(psn,
					       flow->flow_state.r_next_psn);
				if (diff > 0) {
					/* Drop the packet.*/
					goto s_unlock;
				} else if (diff < 0) {
					/*
					 * If a response packet for a restarted
					 * request has come back, reset the
					 * restart flag.
					 */
					if (qp->r_flags & RVT_R_RDMAR_SEQ)
						qp->r_flags &=
							~RVT_R_RDMAR_SEQ;

					/* Drop the packet.*/
					goto s_unlock;
				}

				/*
				 * If SW PSN verification is successful and
				 * this is the last packet in the segment, tell
				 * the caller to process it as a normal packet.
				 */
				fpsn = full_flow_psn(flow,
						     flow->flow_state.lpsn);
				if (cmp_psn(fpsn, psn) == 0) {
					ret = false;
					if (qp->r_flags & RVT_R_RDMAR_SEQ)
						qp->r_flags &=
							~RVT_R_RDMAR_SEQ;
				}
				flow->flow_state.r_next_psn =
					mask_psn(psn + 1);
			} else {
				u32 last_psn;

				last_psn = read_r_next_psn(dd, rcd->ctxt,
							   flow->idx);
				flow->flow_state.r_next_psn = last_psn;
				priv->s_flags |= HFI1_R_TID_SW_PSN;
				/*
				 * If no request has been restarted yet,
				 * restart the current one.
				 */
				if (!(qp->r_flags & RVT_R_RDMAR_SEQ))
					restart_tid_rdma_read_req(rcd, qp,
								  wqe);
			}

			break;

		case RHF_RTE_EXPECTED_FLOW_GEN_ERR:
			/*
			 * Since the TID flow is able to ride through
			 * generation mismatch, drop this stale packet.
			 */
			break;

		default:
			break;
		}
		break;

	case RHF_RCV_TYPE_ERROR:
		switch (rte) {
		case RHF_RTE_ERROR_OP_CODE_ERR:
		case RHF_RTE_ERROR_KHDR_MIN_LEN_ERR:
		case RHF_RTE_ERROR_KHDR_HCRC_ERR:
		case RHF_RTE_ERROR_KHDR_KVER_ERR:
		case RHF_RTE_ERROR_CONTEXT_ERR:
		case RHF_RTE_ERROR_KHDR_TID_ERR:
		default:
			break;
		}
	default:
		break;
	}
s_unlock:
	spin_unlock(&qp->s_lock);
	return ret;
}

bool hfi1_handle_kdeth_eflags(struct hfi1_ctxtdata *rcd,
			      struct hfi1_pportdata *ppd,
			      struct hfi1_packet *packet)
{
	struct hfi1_ibport *ibp = &ppd->ibport_data;
	struct hfi1_devdata *dd = ppd->dd;
	struct rvt_dev_info *rdi = &dd->verbs_dev.rdi;
	u8 rcv_type = rhf_rcv_type(packet->rhf);
	u8 rte = rhf_rcv_type_err(packet->rhf);
	struct ib_header *hdr = packet->hdr;
	struct ib_other_headers *ohdr = NULL;
	int lnh = be16_to_cpu(hdr->lrh[0]) & 3;
	u16 lid  = be16_to_cpu(hdr->lrh[1]);
	u8 opcode;
	u32 qp_num, psn, ibpsn;
	struct rvt_qp *qp;
	struct hfi1_qp_priv *qpriv;
	unsigned long flags;
	bool ret = true;
	struct rvt_ack_entry *e;
	struct tid_rdma_request *req;
	struct tid_rdma_flow *flow;
	int diff = 0;

	trace_hfi1_msg_handle_kdeth_eflags(NULL, "Kdeth error: rhf ",
					   packet->rhf);
	if (packet->rhf & RHF_ICRC_ERR)
		return ret;

	packet->ohdr = &hdr->u.oth;
	ohdr = packet->ohdr;
	trace_input_ibhdr(rcd->dd, packet, !!(rhf_dc_info(packet->rhf)));

	/* Get the destination QP number. */
	qp_num = be32_to_cpu(ohdr->u.tid_rdma.r_rsp.verbs_qp) &
		RVT_QPN_MASK;
	if (lid >= be16_to_cpu(IB_MULTICAST_LID_BASE))
		goto drop;

	psn = mask_psn(be32_to_cpu(ohdr->bth[2]));
	opcode = (be32_to_cpu(ohdr->bth[0]) >> 24) & 0xff;

	rcu_read_lock();
	qp = rvt_lookup_qpn(rdi, &ibp->rvp, qp_num);
	if (!qp)
		goto rcu_unlock;

	packet->qp = qp;

	/* Check for valid receive state. */
	spin_lock_irqsave(&qp->r_lock, flags);
	if (!(ib_rvt_state_ops[qp->state] & RVT_PROCESS_RECV_OK)) {
		ibp->rvp.n_pkt_drops++;
		goto r_unlock;
	}

	if (packet->rhf & RHF_TID_ERR) {
		/* For TIDERR and RC QPs preemptively schedule a NAK */
		u32 tlen = rhf_pkt_len(packet->rhf); /* in bytes */

		/* Sanity check packet */
		if (tlen < 24)
			goto r_unlock;

		/*
		 * Check for GRH. We should never get packets with GRH in this
		 * path.
		 */
		if (lnh == HFI1_LRH_GRH)
			goto r_unlock;

		if (tid_rdma_tid_err(packet, rcv_type))
			goto r_unlock;
	}

	/* handle TID RDMA READ */
	if (opcode == TID_OP(READ_RESP)) {
		ibpsn = be32_to_cpu(ohdr->u.tid_rdma.r_rsp.verbs_psn);
		ibpsn = mask_psn(ibpsn);
		ret = handle_read_kdeth_eflags(rcd, packet, rcv_type, rte, psn,
					       ibpsn);
		goto r_unlock;
	}

	/*
	 * qp->s_tail_ack_queue points to the rvt_ack_entry currently being
	 * processed. These a completed sequentially so we can be sure that
	 * the pointer will not change until the entire request has completed.
	 */
	spin_lock(&qp->s_lock);
	qpriv = qp->priv;
	if (qpriv->r_tid_tail == HFI1_QP_WQE_INVALID ||
	    qpriv->r_tid_tail == qpriv->r_tid_head)
		goto unlock;
	e = &qp->s_ack_queue[qpriv->r_tid_tail];
	if (e->opcode != TID_OP(WRITE_REQ))
		goto unlock;
	req = ack_to_tid_req(e);
	if (req->comp_seg == req->cur_seg)
		goto unlock;
	flow = &req->flows[req->clear_tail];
	trace_hfi1_eflags_err_write(qp, rcv_type, rte, psn);
	trace_hfi1_rsp_handle_kdeth_eflags(qp, psn);
	trace_hfi1_tid_write_rsp_handle_kdeth_eflags(qp);
	trace_hfi1_tid_req_handle_kdeth_eflags(qp, 0, e->opcode, e->psn,
					       e->lpsn, req);
	trace_hfi1_tid_flow_handle_kdeth_eflags(qp, req->clear_tail, flow);

	switch (rcv_type) {
	case RHF_RCV_TYPE_EXPECTED:
		switch (rte) {
		case RHF_RTE_EXPECTED_FLOW_SEQ_ERR:
			if (!(qpriv->s_flags & HFI1_R_TID_SW_PSN)) {
				qpriv->s_flags |= HFI1_R_TID_SW_PSN;
				flow->flow_state.r_next_psn =
					read_r_next_psn(dd, rcd->ctxt,
							flow->idx);
				qpriv->r_next_psn_kdeth =
					flow->flow_state.r_next_psn;
				goto nak_psn;
			} else {
				/*
				 * If the received PSN does not match the next
				 * expected PSN, NAK the packet.
				 * However, only do that if we know that the a
				 * NAK has already been sent. Otherwise, this
				 * mismatch could be due to packets that were
				 * already in flight.
				 */
				diff = cmp_psn(psn,
					       flow->flow_state.r_next_psn);
				if (diff > 0)
					goto nak_psn;
				else if (diff < 0)
					break;

				qpriv->s_nak_state = 0;
				/*
				 * If SW PSN verification is successful and this
				 * is the last packet in the segment, tell the
				 * caller to process it as a normal packet.
				 */
				if (psn == full_flow_psn(flow,
							 flow->flow_state.lpsn))
					ret = false;
				flow->flow_state.r_next_psn =
					mask_psn(psn + 1);
				qpriv->r_next_psn_kdeth =
					flow->flow_state.r_next_psn;
			}
			break;

		case RHF_RTE_EXPECTED_FLOW_GEN_ERR:
			goto nak_psn;

		default:
			break;
		}
		break;

	case RHF_RCV_TYPE_ERROR:
		switch (rte) {
		case RHF_RTE_ERROR_OP_CODE_ERR:
		case RHF_RTE_ERROR_KHDR_MIN_LEN_ERR:
		case RHF_RTE_ERROR_KHDR_HCRC_ERR:
		case RHF_RTE_ERROR_KHDR_KVER_ERR:
		case RHF_RTE_ERROR_CONTEXT_ERR:
		case RHF_RTE_ERROR_KHDR_TID_ERR:
		default:
			break;
		}
	default:
		break;
	}

unlock:
	spin_unlock(&qp->s_lock);
r_unlock:
	spin_unlock_irqrestore(&qp->r_lock, flags);
rcu_unlock:
	rcu_read_unlock();
drop:
	return ret;
nak_psn:
	ibp->rvp.n_rc_seqnak++;
	if (!qpriv->s_nak_state) {
		qpriv->s_nak_state = IB_NAK_PSN_ERROR;
		/* We are NAK'ing the next expected PSN */
		qpriv->s_nak_psn = mask_psn(flow->flow_state.r_next_psn);
		qpriv->s_flags |= RVT_S_ACK_PENDING;
		if (qpriv->r_tid_ack == HFI1_QP_WQE_INVALID)
			qpriv->r_tid_ack = qpriv->r_tid_tail;
		hfi1_schedule_tid_send(qp);
	}
	goto unlock;
}

/*
 * "Rewind" the TID request information.
 * This means that we reset the state back to ACTIVE,
 * find the proper flow, set the flow index to that flow,
 * and reset the flow information.
 */
void hfi1_tid_rdma_restart_req(struct rvt_qp *qp, struct rvt_swqe *wqe,
			       u32 *bth2)
{
	struct tid_rdma_request *req = wqe_to_tid_req(wqe);
	struct tid_rdma_flow *flow;
	struct hfi1_qp_priv *qpriv = qp->priv;
	int diff, delta_pkts;
	u32 tididx = 0, i;
	u16 fidx;

	if (wqe->wr.opcode == IB_WR_TID_RDMA_READ) {
		*bth2 = mask_psn(qp->s_psn);
		flow = find_flow_ib(req, *bth2, &fidx);
		if (!flow) {
			trace_hfi1_msg_tid_restart_req(/* msg */
			   qp, "!!!!!! Could not find flow to restart: bth2 ",
			   (u64)*bth2);
			trace_hfi1_tid_req_restart_req(qp, 0, wqe->wr.opcode,
						       wqe->psn, wqe->lpsn,
						       req);
			return;
		}
	} else {
		fidx = req->acked_tail;
		flow = &req->flows[fidx];
		*bth2 = mask_psn(req->r_ack_psn);
	}

	if (wqe->wr.opcode == IB_WR_TID_RDMA_READ)
		delta_pkts = delta_psn(*bth2, flow->flow_state.ib_spsn);
	else
		delta_pkts = delta_psn(*bth2,
				       full_flow_psn(flow,
						     flow->flow_state.spsn));

	trace_hfi1_tid_flow_restart_req(qp, fidx, flow);
	diff = delta_pkts + flow->resync_npkts;

	flow->sent = 0;
	flow->pkt = 0;
	flow->tid_idx = 0;
	flow->tid_offset = 0;
	if (diff) {
		for (tididx = 0; tididx < flow->tidcnt; tididx++) {
			u32 tidentry = flow->tid_entry[tididx], tidlen,
				tidnpkts, npkts;

			flow->tid_offset = 0;
			tidlen = EXP_TID_GET(tidentry, LEN) * PAGE_SIZE;
			tidnpkts = rvt_div_round_up_mtu(qp, tidlen);
			npkts = min_t(u32, diff, tidnpkts);
			flow->pkt += npkts;
			flow->sent += (npkts == tidnpkts ? tidlen :
				       npkts * qp->pmtu);
			flow->tid_offset += npkts * qp->pmtu;
			diff -= npkts;
			if (!diff)
				break;
		}
	}
	if (wqe->wr.opcode == IB_WR_TID_RDMA_WRITE) {
		rvt_skip_sge(&qpriv->tid_ss, (req->cur_seg * req->seg_len) +
			     flow->sent, 0);
		/*
		 * Packet PSN is based on flow_state.spsn + flow->pkt. However,
		 * during a RESYNC, the generation is incremented and the
		 * sequence is reset to 0. Since we've adjusted the npkts in the
		 * flow and the SGE has been sufficiently advanced, we have to
		 * adjust flow->pkt in order to calculate the correct PSN.
		 */
		flow->pkt -= flow->resync_npkts;
	}

	if (flow->tid_offset ==
	    EXP_TID_GET(flow->tid_entry[tididx], LEN) * PAGE_SIZE) {
		tididx++;
		flow->tid_offset = 0;
	}
	flow->tid_idx = tididx;
	if (wqe->wr.opcode == IB_WR_TID_RDMA_READ)
		/* Move flow_idx to correct index */
		req->flow_idx = fidx;
	else
		req->clear_tail = fidx;

	trace_hfi1_tid_flow_restart_req(qp, fidx, flow);
	trace_hfi1_tid_req_restart_req(qp, 0, wqe->wr.opcode, wqe->psn,
				       wqe->lpsn, req);
	req->state = TID_REQUEST_ACTIVE;
	if (wqe->wr.opcode == IB_WR_TID_RDMA_WRITE) {
		/* Reset all the flows that we are going to resend */
		fidx = CIRC_NEXT(fidx, MAX_FLOWS);
		i = qpriv->s_tid_tail;
		do {
			for (; CIRC_CNT(req->setup_head, fidx, MAX_FLOWS);
			      fidx = CIRC_NEXT(fidx, MAX_FLOWS)) {
				req->flows[fidx].sent = 0;
				req->flows[fidx].pkt = 0;
				req->flows[fidx].tid_idx = 0;
				req->flows[fidx].tid_offset = 0;
				req->flows[fidx].resync_npkts = 0;
			}
			if (i == qpriv->s_tid_cur)
				break;
			do {
				i = (++i == qp->s_size ? 0 : i);
				wqe = rvt_get_swqe_ptr(qp, i);
			} while (wqe->wr.opcode != IB_WR_TID_RDMA_WRITE);
			req = wqe_to_tid_req(wqe);
			req->cur_seg = req->ack_seg;
			fidx = req->acked_tail;
			/* Pull req->clear_tail back */
			req->clear_tail = fidx;
		} while (1);
	}
}

void hfi1_qp_kern_exp_rcv_clear_all(struct rvt_qp *qp)
{
	int i, ret;
	struct hfi1_qp_priv *qpriv = qp->priv;
	struct tid_flow_state *fs;

	if (qp->ibqp.qp_type != IB_QPT_RC || !HFI1_CAP_IS_KSET(TID_RDMA))
		return;

	/*
	 * First, clear the flow to help prevent any delayed packets from
	 * being delivered.
	 */
	fs = &qpriv->flow_state;
	if (fs->index != RXE_NUM_TID_FLOWS)
		hfi1_kern_clear_hw_flow(qpriv->rcd, qp);

	for (i = qp->s_acked; i != qp->s_head;) {
		struct rvt_swqe *wqe = rvt_get_swqe_ptr(qp, i);

		if (++i == qp->s_size)
			i = 0;
		/* Free only locally allocated TID entries */
		if (wqe->wr.opcode != IB_WR_TID_RDMA_READ)
			continue;
		do {
			struct hfi1_swqe_priv *priv = wqe->priv;

			ret = hfi1_kern_exp_rcv_clear(&priv->tid_req);
		} while (!ret);
	}
	for (i = qp->s_acked_ack_queue; i != qp->r_head_ack_queue;) {
		struct rvt_ack_entry *e = &qp->s_ack_queue[i];

		if (++i == rvt_max_atomic(ib_to_rvt(qp->ibqp.device)))
			i = 0;
		/* Free only locally allocated TID entries */
		if (e->opcode != TID_OP(WRITE_REQ))
			continue;
		do {
			struct hfi1_ack_priv *priv = e->priv;

			ret = hfi1_kern_exp_rcv_clear(&priv->tid_req);
		} while (!ret);
	}
}

bool hfi1_tid_rdma_wqe_interlock(struct rvt_qp *qp, struct rvt_swqe *wqe)
{
	struct rvt_swqe *prev;
	struct hfi1_qp_priv *priv = qp->priv;
	u32 s_prev;
	struct tid_rdma_request *req;

	s_prev = (qp->s_cur == 0 ? qp->s_size : qp->s_cur) - 1;
	prev = rvt_get_swqe_ptr(qp, s_prev);

	switch (wqe->wr.opcode) {
	case IB_WR_SEND:
	case IB_WR_SEND_WITH_IMM:
	case IB_WR_SEND_WITH_INV:
	case IB_WR_ATOMIC_CMP_AND_SWP:
	case IB_WR_ATOMIC_FETCH_AND_ADD:
	case IB_WR_RDMA_WRITE:
		switch (prev->wr.opcode) {
		case IB_WR_TID_RDMA_WRITE:
			req = wqe_to_tid_req(prev);
			if (req->ack_seg != req->total_segs)
				goto interlock;
		default:
			break;
		}
		break;
	case IB_WR_RDMA_READ:
		if (prev->wr.opcode != IB_WR_TID_RDMA_WRITE)
			break;
		/* fall through */
	case IB_WR_TID_RDMA_READ:
		switch (prev->wr.opcode) {
		case IB_WR_RDMA_READ:
			if (qp->s_acked != qp->s_cur)
				goto interlock;
			break;
		case IB_WR_TID_RDMA_WRITE:
			req = wqe_to_tid_req(prev);
			if (req->ack_seg != req->total_segs)
				goto interlock;
		default:
			break;
		}
	default:
		break;
	}
	return false;

interlock:
	priv->s_flags |= HFI1_S_TID_WAIT_INTERLCK;
	return true;
}

/* Does @sge meet the alignment requirements for tid rdma? */
static inline bool hfi1_check_sge_align(struct rvt_qp *qp,
					struct rvt_sge *sge, int num_sge)
{
	int i;

	for (i = 0; i < num_sge; i++, sge++) {
		trace_hfi1_sge_check_align(qp, i, sge);
		if ((u64)sge->vaddr & ~PAGE_MASK ||
		    sge->sge_length & ~PAGE_MASK)
			return false;
	}
	return true;
}

void setup_tid_rdma_wqe(struct rvt_qp *qp, struct rvt_swqe *wqe)
{
	struct hfi1_qp_priv *qpriv = (struct hfi1_qp_priv *)qp->priv;
	struct hfi1_swqe_priv *priv = wqe->priv;
	struct tid_rdma_params *remote;
	enum ib_wr_opcode new_opcode;
	bool do_tid_rdma = false;
	struct hfi1_pportdata *ppd = qpriv->rcd->ppd;

	if ((rdma_ah_get_dlid(&qp->remote_ah_attr) & ~((1 << ppd->lmc) - 1)) ==
				ppd->lid)
		return;
	if (qpriv->hdr_type != HFI1_PKT_TYPE_9B)
		return;

	rcu_read_lock();
	remote = rcu_dereference(qpriv->tid_rdma.remote);
	/*
	 * If TID RDMA is disabled by the negotiation, don't
	 * use it.
	 */
	if (!remote)
		goto exit;

	if (wqe->wr.opcode == IB_WR_RDMA_READ) {
		if (hfi1_check_sge_align(qp, &wqe->sg_list[0],
					 wqe->wr.num_sge)) {
			new_opcode = IB_WR_TID_RDMA_READ;
			do_tid_rdma = true;
		}
	} else if (wqe->wr.opcode == IB_WR_RDMA_WRITE) {
		/*
		 * TID RDMA is enabled for this RDMA WRITE request iff:
		 *   1. The remote address is page-aligned,
		 *   2. The length is larger than the minimum segment size,
		 *   3. The length is page-multiple.
		 */
		if (!(wqe->rdma_wr.remote_addr & ~PAGE_MASK) &&
		    !(wqe->length & ~PAGE_MASK)) {
			new_opcode = IB_WR_TID_RDMA_WRITE;
			do_tid_rdma = true;
		}
	}

	if (do_tid_rdma) {
		if (hfi1_kern_exp_rcv_alloc_flows(&priv->tid_req, GFP_ATOMIC))
			goto exit;
		wqe->wr.opcode = new_opcode;
		priv->tid_req.seg_len =
			min_t(u32, remote->max_len, wqe->length);
		priv->tid_req.total_segs =
			DIV_ROUND_UP(wqe->length, priv->tid_req.seg_len);
		/* Compute the last PSN of the request */
		wqe->lpsn = wqe->psn;
		if (wqe->wr.opcode == IB_WR_TID_RDMA_READ) {
			priv->tid_req.n_flows = remote->max_read;
			qpriv->tid_r_reqs++;
			wqe->lpsn += rvt_div_round_up_mtu(qp, wqe->length) - 1;
		} else {
			wqe->lpsn += priv->tid_req.total_segs - 1;
			atomic_inc(&qpriv->n_requests);
		}

		priv->tid_req.cur_seg = 0;
		priv->tid_req.comp_seg = 0;
		priv->tid_req.ack_seg = 0;
		priv->tid_req.state = TID_REQUEST_INACTIVE;
		/*
		 * Reset acked_tail.
		 * TID RDMA READ does not have ACKs so it does not
		 * update the pointer. We have to reset it so TID RDMA
		 * WRITE does not get confused.
		 */
		priv->tid_req.acked_tail = priv->tid_req.setup_head;
		trace_hfi1_tid_req_setup_tid_wqe(qp, 1, wqe->wr.opcode,
						 wqe->psn, wqe->lpsn,
						 &priv->tid_req);
	}
exit:
	rcu_read_unlock();
}

/* TID RDMA WRITE functions */

u32 hfi1_build_tid_rdma_write_req(struct rvt_qp *qp, struct rvt_swqe *wqe,
				  struct ib_other_headers *ohdr,
				  u32 *bth1, u32 *bth2, u32 *len)
{
	struct hfi1_qp_priv *qpriv = qp->priv;
	struct tid_rdma_request *req = wqe_to_tid_req(wqe);
	struct tid_rdma_params *remote;

	rcu_read_lock();
	remote = rcu_dereference(qpriv->tid_rdma.remote);
	/*
	 * Set the number of flow to be used based on negotiated
	 * parameters.
	 */
	req->n_flows = remote->max_write;
	req->state = TID_REQUEST_ACTIVE;

	KDETH_RESET(ohdr->u.tid_rdma.w_req.kdeth0, KVER, 0x1);
	KDETH_RESET(ohdr->u.tid_rdma.w_req.kdeth1, JKEY, remote->jkey);
	ohdr->u.tid_rdma.w_req.reth.vaddr =
		cpu_to_be64(wqe->rdma_wr.remote_addr + (wqe->length - *len));
	ohdr->u.tid_rdma.w_req.reth.rkey =
		cpu_to_be32(wqe->rdma_wr.rkey);
	ohdr->u.tid_rdma.w_req.reth.length = cpu_to_be32(*len);
	ohdr->u.tid_rdma.w_req.verbs_qp = cpu_to_be32(qp->remote_qpn);
	*bth1 &= ~RVT_QPN_MASK;
	*bth1 |= remote->qp;
	qp->s_state = TID_OP(WRITE_REQ);
	qp->s_flags |= HFI1_S_WAIT_TID_RESP;
	*bth2 |= IB_BTH_REQ_ACK;
	*len = 0;

	rcu_read_unlock();
	return sizeof(ohdr->u.tid_rdma.w_req) / sizeof(u32);
}

void hfi1_compute_tid_rdma_flow_wt(void)
{
	/*
	 * Heuristic for computing the RNR timeout when waiting on the flow
	 * queue. Rather than a computationaly expensive exact estimate of when
	 * a flow will be available, we assume that if a QP is at position N in
	 * the flow queue it has to wait approximately (N + 1) * (number of
	 * segments between two sync points), assuming PMTU of 4K. The rationale
	 * for this is that flows are released and recycled at each sync point.
	 */
	tid_rdma_flow_wt = MAX_TID_FLOW_PSN * enum_to_mtu(OPA_MTU_4096) /
		TID_RDMA_MAX_SEGMENT_SIZE;
}

static u32 position_in_queue(struct hfi1_qp_priv *qpriv,
			     struct tid_queue *queue)
{
	return qpriv->tid_enqueue - queue->dequeue;
}

/*
 * @qp: points to rvt_qp context.
 * @to_seg: desired RNR timeout in segments.
 * Return: index of the next highest timeout in the ib_hfi1_rnr_table[]
 */
static u32 hfi1_compute_tid_rnr_timeout(struct rvt_qp *qp, u32 to_seg)
{
	struct hfi1_qp_priv *qpriv = qp->priv;
	u64 timeout;
	u32 bytes_per_us;
	u8 i;

	bytes_per_us = active_egress_rate(qpriv->rcd->ppd) / 8;
	timeout = (to_seg * TID_RDMA_MAX_SEGMENT_SIZE) / bytes_per_us;
	/*
	 * Find the next highest value in the RNR table to the required
	 * timeout. This gives the responder some padding.
	 */
	for (i = 1; i <= IB_AETH_CREDIT_MASK; i++)
		if (rvt_rnr_tbl_to_usec(i) >= timeout)
			return i;
	return 0;
}

/**
 * Central place for resource allocation at TID write responder,
 * is called from write_req and write_data interrupt handlers as
 * well as the send thread when a queued QP is scheduled for
 * resource allocation.
 *
 * Iterates over (a) segments of a request and then (b) queued requests
 * themselves to allocate resources for up to local->max_write
 * segments across multiple requests. Stop allocating when we
 * hit a sync point, resume allocating after data packets at
 * sync point have been received.
 *
 * Resource allocation and sending of responses is decoupled. The
 * request/segment which are being allocated and sent are as follows.
 * Resources are allocated for:
 *     [request: qpriv->r_tid_alloc, segment: req->alloc_seg]
 * The send thread sends:
 *     [request: qp->s_tail_ack_queue, segment:req->cur_seg]
 */
static void hfi1_tid_write_alloc_resources(struct rvt_qp *qp, bool intr_ctx)
{
	struct tid_rdma_request *req;
	struct hfi1_qp_priv *qpriv = qp->priv;
	struct hfi1_ctxtdata *rcd = qpriv->rcd;
	struct tid_rdma_params *local = &qpriv->tid_rdma.local;
	struct rvt_ack_entry *e;
	u32 npkts, to_seg;
	bool last;
	int ret = 0;

	lockdep_assert_held(&qp->s_lock);

	while (1) {
		trace_hfi1_rsp_tid_write_alloc_res(qp, 0);
		trace_hfi1_tid_write_rsp_alloc_res(qp);
		/*
		 * Don't allocate more segments if a RNR NAK has already been
		 * scheduled to avoid messing up qp->r_psn: the RNR NAK will
		 * be sent only when all allocated segments have been sent.
		 * However, if more segments are allocated before that, TID RDMA
		 * WRITE RESP packets will be sent out for these new segments
		 * before the RNR NAK packet. When the requester receives the
		 * RNR NAK packet, it will restart with qp->s_last_psn + 1,
		 * which does not match qp->r_psn and will be dropped.
		 * Consequently, the requester will exhaust its retries and
		 * put the qp into error state.
		 */
		if (qpriv->rnr_nak_state == TID_RNR_NAK_SEND)
			break;

		/* No requests left to process */
		if (qpriv->r_tid_alloc == qpriv->r_tid_head) {
			/* If all data has been received, clear the flow */
			if (qpriv->flow_state.index < RXE_NUM_TID_FLOWS &&
			    !qpriv->alloc_w_segs) {
				hfi1_kern_clear_hw_flow(rcd, qp);
				qpriv->s_flags &= ~HFI1_R_TID_SW_PSN;
			}
			break;
		}

		e = &qp->s_ack_queue[qpriv->r_tid_alloc];
		if (e->opcode != TID_OP(WRITE_REQ))
			goto next_req;
		req = ack_to_tid_req(e);
		trace_hfi1_tid_req_write_alloc_res(qp, 0, e->opcode, e->psn,
						   e->lpsn, req);
		/* Finished allocating for all segments of this request */
		if (req->alloc_seg >= req->total_segs)
			goto next_req;

		/* Can allocate only a maximum of local->max_write for a QP */
		if (qpriv->alloc_w_segs >= local->max_write)
			break;

		/* Don't allocate at a sync point with data packets pending */
		if (qpriv->sync_pt && qpriv->alloc_w_segs)
			break;

		/* All data received at the sync point, continue */
		if (qpriv->sync_pt && !qpriv->alloc_w_segs) {
			hfi1_kern_clear_hw_flow(rcd, qp);
			qpriv->sync_pt = false;
			qpriv->s_flags &= ~HFI1_R_TID_SW_PSN;
		}

		/* Allocate flow if we don't have one */
		if (qpriv->flow_state.index >= RXE_NUM_TID_FLOWS) {
			ret = hfi1_kern_setup_hw_flow(qpriv->rcd, qp);
			if (ret) {
				to_seg = tid_rdma_flow_wt *
					position_in_queue(qpriv,
							  &rcd->flow_queue);
				break;
			}
		}

		npkts = rvt_div_round_up_mtu(qp, req->seg_len);

		/*
		 * We are at a sync point if we run out of KDETH PSN space.
		 * Last PSN of every generation is reserved for RESYNC.
		 */
		if (qpriv->flow_state.psn + npkts > MAX_TID_FLOW_PSN - 1) {
			qpriv->sync_pt = true;
			break;
		}

		/*
		 * If overtaking req->acked_tail, send an RNR NAK. Because the
		 * QP is not queued in this case, and the issue can only be
		 * caused due a delay in scheduling the second leg which we
		 * cannot estimate, we use a rather arbitrary RNR timeout of
		 * (MAX_FLOWS / 2) segments
		 */
		if (!CIRC_SPACE(req->setup_head, req->acked_tail,
				MAX_FLOWS)) {
			ret = -EAGAIN;
			to_seg = MAX_FLOWS >> 1;
			qpriv->s_flags |= RVT_S_ACK_PENDING;
			hfi1_schedule_tid_send(qp);
			break;
		}

		/* Try to allocate rcv array / TID entries */
		ret = hfi1_kern_exp_rcv_setup(req, &req->ss, &last);
		if (ret == -EAGAIN)
			to_seg = position_in_queue(qpriv, &rcd->rarr_queue);
		if (ret)
			break;

		qpriv->alloc_w_segs++;
		req->alloc_seg++;
		continue;
next_req:
		/* Begin processing the next request */
		if (++qpriv->r_tid_alloc >
		    rvt_size_atomic(ib_to_rvt(qp->ibqp.device)))
			qpriv->r_tid_alloc = 0;
	}

	/*
	 * Schedule an RNR NAK to be sent if (a) flow or rcv array allocation
	 * has failed (b) we are called from the rcv handler interrupt context
	 * (c) an RNR NAK has not already been scheduled
	 */
	if (ret == -EAGAIN && intr_ctx && !qp->r_nak_state)
		goto send_rnr_nak;

	return;

send_rnr_nak:
	lockdep_assert_held(&qp->r_lock);

	/* Set r_nak_state to prevent unrelated events from generating NAK's */
	qp->r_nak_state = hfi1_compute_tid_rnr_timeout(qp, to_seg) | IB_RNR_NAK;

	/* Pull back r_psn to the segment being RNR NAK'd */
	qp->r_psn = e->psn + req->alloc_seg;
	qp->r_ack_psn = qp->r_psn;
	/*
	 * Pull back r_head_ack_queue to the ack entry following the request
	 * being RNR NAK'd. This allows resources to be allocated to the request
	 * if the queued QP is scheduled.
	 */
	qp->r_head_ack_queue = qpriv->r_tid_alloc + 1;
	if (qp->r_head_ack_queue > rvt_size_atomic(ib_to_rvt(qp->ibqp.device)))
		qp->r_head_ack_queue = 0;
	qpriv->r_tid_head = qp->r_head_ack_queue;
	/*
	 * These send side fields are used in make_rc_ack(). They are set in
	 * hfi1_send_rc_ack() but must be set here before dropping qp->s_lock
	 * for consistency
	 */
	qp->s_nak_state = qp->r_nak_state;
	qp->s_ack_psn = qp->r_ack_psn;
	/*
	 * Clear the ACK PENDING flag to prevent unwanted ACK because we
	 * have modified qp->s_ack_psn here.
	 */
	qp->s_flags &= ~(RVT_S_ACK_PENDING);

	trace_hfi1_rsp_tid_write_alloc_res(qp, qp->r_psn);
	/*
	 * qpriv->rnr_nak_state is used to determine when the scheduled RNR NAK
	 * has actually been sent. qp->s_flags RVT_S_ACK_PENDING bit cannot be
	 * used for this because qp->s_lock is dropped before calling
	 * hfi1_send_rc_ack() leading to inconsistency between the receive
	 * interrupt handlers and the send thread in make_rc_ack()
	 */
	qpriv->rnr_nak_state = TID_RNR_NAK_SEND;

	/*
	 * Schedule RNR NAK to be sent. RNR NAK's are scheduled from the receive
	 * interrupt handlers but will be sent from the send engine behind any
	 * previous responses that may have been scheduled
	 */
	rc_defered_ack(rcd, qp);
}

void hfi1_rc_rcv_tid_rdma_write_req(struct hfi1_packet *packet)
{
	/* HANDLER FOR TID RDMA WRITE REQUEST packet (Responder side)*/

	/*
	 * 1. Verify TID RDMA WRITE REQ as per IB_OPCODE_RC_RDMA_WRITE_FIRST
	 *    (see hfi1_rc_rcv())
	 *     - Don't allow 0-length requests.
	 * 2. Put TID RDMA WRITE REQ into the response queueu (s_ack_queue)
	 *     - Setup struct tid_rdma_req with request info
	 *     - Prepare struct tid_rdma_flow array?
	 * 3. Set the qp->s_ack_state as state diagram in design doc.
	 * 4. Set RVT_S_RESP_PENDING in s_flags.
	 * 5. Kick the send engine (hfi1_schedule_send())
	 */
	struct hfi1_ctxtdata *rcd = packet->rcd;
	struct rvt_qp *qp = packet->qp;
	struct hfi1_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
	struct ib_other_headers *ohdr = packet->ohdr;
	struct rvt_ack_entry *e;
	unsigned long flags;
	struct ib_reth *reth;
	struct hfi1_qp_priv *qpriv = qp->priv;
	struct tid_rdma_request *req;
	u32 bth0, psn, len, rkey, num_segs;
	bool fecn;
	u8 next;
	u64 vaddr;
	int diff;

	bth0 = be32_to_cpu(ohdr->bth[0]);
	if (hfi1_ruc_check_hdr(ibp, packet))
		return;

	fecn = process_ecn(qp, packet);
	psn = mask_psn(be32_to_cpu(ohdr->bth[2]));
	trace_hfi1_rsp_rcv_tid_write_req(qp, psn);

	if (qp->state == IB_QPS_RTR && !(qp->r_flags & RVT_R_COMM_EST))
		rvt_comm_est(qp);

	if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_WRITE)))
		goto nack_inv;

	reth = &ohdr->u.tid_rdma.w_req.reth;
	vaddr = be64_to_cpu(reth->vaddr);
	len = be32_to_cpu(reth->length);

	num_segs = DIV_ROUND_UP(len, qpriv->tid_rdma.local.max_len);
	diff = delta_psn(psn, qp->r_psn);
	if (unlikely(diff)) {
		tid_rdma_rcv_err(packet, ohdr, qp, psn, diff, fecn);
		return;
	}

	/*
	 * The resent request which was previously RNR NAK'd is inserted at the
	 * location of the original request, which is one entry behind
	 * r_head_ack_queue
	 */
	if (qpriv->rnr_nak_state)
		qp->r_head_ack_queue = qp->r_head_ack_queue ?
			qp->r_head_ack_queue - 1 :
			rvt_size_atomic(ib_to_rvt(qp->ibqp.device));

	/* We've verified the request, insert it into the ack queue. */
	next = qp->r_head_ack_queue + 1;
	if (next > rvt_size_atomic(ib_to_rvt(qp->ibqp.device)))
		next = 0;
	spin_lock_irqsave(&qp->s_lock, flags);
	if (unlikely(next == qp->s_acked_ack_queue)) {
		if (!qp->s_ack_queue[next].sent)
			goto nack_inv_unlock;
		update_ack_queue(qp, next);
	}
	e = &qp->s_ack_queue[qp->r_head_ack_queue];
	req = ack_to_tid_req(e);

	/* Bring previously RNR NAK'd request back to life */
	if (qpriv->rnr_nak_state) {
		qp->r_nak_state = 0;
		qp->s_nak_state = 0;
		qpriv->rnr_nak_state = TID_RNR_NAK_INIT;
		qp->r_psn = e->lpsn + 1;
		req->state = TID_REQUEST_INIT;
		goto update_head;
	}

	release_rdma_sge_mr(e);

	/* The length needs to be in multiples of PAGE_SIZE */
	if (!len || len & ~PAGE_MASK)
		goto nack_inv_unlock;

	rkey = be32_to_cpu(reth->rkey);
	qp->r_len = len;

	if (e->opcode == TID_OP(WRITE_REQ) &&
	    (req->setup_head != req->clear_tail ||
	     req->clear_tail != req->acked_tail))
		goto nack_inv_unlock;

	if (unlikely(!rvt_rkey_ok(qp, &e->rdma_sge, qp->r_len, vaddr,
				  rkey, IB_ACCESS_REMOTE_WRITE)))
		goto nack_acc;

	qp->r_psn += num_segs - 1;

	e->opcode = (bth0 >> 24) & 0xff;
	e->psn = psn;
	e->lpsn = qp->r_psn;
	e->sent = 0;

	req->n_flows = min_t(u16, num_segs, qpriv->tid_rdma.local.max_write);
	req->state = TID_REQUEST_INIT;
	req->cur_seg = 0;
	req->comp_seg = 0;
	req->ack_seg = 0;
	req->alloc_seg = 0;
	req->isge = 0;
	req->seg_len = qpriv->tid_rdma.local.max_len;
	req->total_len = len;
	req->total_segs = num_segs;
	req->r_flow_psn = e->psn;
	req->ss.sge = e->rdma_sge;
	req->ss.num_sge = 1;

	req->flow_idx = req->setup_head;
	req->clear_tail = req->setup_head;
	req->acked_tail = req->setup_head;

	qp->r_state = e->opcode;
	qp->r_nak_state = 0;
	/*
	 * We need to increment the MSN here instead of when we
	 * finish sending the result since a duplicate request would
	 * increment it more than once.
	 */
	qp->r_msn++;
	qp->r_psn++;

	trace_hfi1_tid_req_rcv_write_req(qp, 0, e->opcode, e->psn, e->lpsn,
					 req);

	if (qpriv->r_tid_tail == HFI1_QP_WQE_INVALID) {
		qpriv->r_tid_tail = qp->r_head_ack_queue;
	} else if (qpriv->r_tid_tail == qpriv->r_tid_head) {
		struct tid_rdma_request *ptr;

		e = &qp->s_ack_queue[qpriv->r_tid_tail];
		ptr = ack_to_tid_req(e);

		if (e->opcode != TID_OP(WRITE_REQ) ||
		    ptr->comp_seg == ptr->total_segs) {
			if (qpriv->r_tid_tail == qpriv->r_tid_ack)
				qpriv->r_tid_ack = qp->r_head_ack_queue;
			qpriv->r_tid_tail = qp->r_head_ack_queue;
		}
	}
update_head:
	qp->r_head_ack_queue = next;
	qpriv->r_tid_head = qp->r_head_ack_queue;

	hfi1_tid_write_alloc_resources(qp, true);
	trace_hfi1_tid_write_rsp_rcv_req(qp);

	/* Schedule the send tasklet. */
	qp->s_flags |= RVT_S_RESP_PENDING;
	if (fecn)
		qp->s_flags |= RVT_S_ECN;
	hfi1_schedule_send(qp);

	spin_unlock_irqrestore(&qp->s_lock, flags);
	return;

nack_inv_unlock:
	spin_unlock_irqrestore(&qp->s_lock, flags);
nack_inv:
	rvt_rc_error(qp, IB_WC_LOC_QP_OP_ERR);
	qp->r_nak_state = IB_NAK_INVALID_REQUEST;
	qp->r_ack_psn = qp->r_psn;
	/* Queue NAK for later */
	rc_defered_ack(rcd, qp);
	return;
nack_acc:
	spin_unlock_irqrestore(&qp->s_lock, flags);
	rvt_rc_error(qp, IB_WC_LOC_PROT_ERR);
	qp->r_nak_state = IB_NAK_REMOTE_ACCESS_ERROR;
	qp->r_ack_psn = qp->r_psn;
}

u32 hfi1_build_tid_rdma_write_resp(struct rvt_qp *qp, struct rvt_ack_entry *e,
				   struct ib_other_headers *ohdr, u32 *bth1,
				   u32 bth2, u32 *len,
				   struct rvt_sge_state **ss)
{
	struct hfi1_ack_priv *epriv = e->priv;
	struct tid_rdma_request *req = &epriv->tid_req;
	struct hfi1_qp_priv *qpriv = qp->priv;
	struct tid_rdma_flow *flow = NULL;
	u32 resp_len = 0, hdwords = 0;
	void *resp_addr = NULL;
	struct tid_rdma_params *remote;

	trace_hfi1_tid_req_build_write_resp(qp, 0, e->opcode, e->psn, e->lpsn,
					    req);
	trace_hfi1_tid_write_rsp_build_resp(qp);
	trace_hfi1_rsp_build_tid_write_resp(qp, bth2);
	flow = &req->flows[req->flow_idx];
	switch (req->state) {
	default:
		/*
		 * Try to allocate resources here in case QP was queued and was
		 * later scheduled when resources became available
		 */
		hfi1_tid_write_alloc_resources(qp, false);

		/* We've already sent everything which is ready */
		if (req->cur_seg >= req->alloc_seg)
			goto done;

		/*
		 * Resources can be assigned but responses cannot be sent in
		 * rnr_nak state, till the resent request is received
		 */
		if (qpriv->rnr_nak_state == TID_RNR_NAK_SENT)
			goto done;

		req->state = TID_REQUEST_ACTIVE;
		trace_hfi1_tid_flow_build_write_resp(qp, req->flow_idx, flow);
		req->flow_idx = CIRC_NEXT(req->flow_idx, MAX_FLOWS);
		hfi1_add_tid_reap_timer(qp);
		break;

	case TID_REQUEST_RESEND_ACTIVE:
	case TID_REQUEST_RESEND:
		trace_hfi1_tid_flow_build_write_resp(qp, req->flow_idx, flow);
		req->flow_idx = CIRC_NEXT(req->flow_idx, MAX_FLOWS);
		if (!CIRC_CNT(req->setup_head, req->flow_idx, MAX_FLOWS))
			req->state = TID_REQUEST_ACTIVE;

		hfi1_mod_tid_reap_timer(qp);
		break;
	}
	flow->flow_state.resp_ib_psn = bth2;
	resp_addr = (void *)flow->tid_entry;
	resp_len = sizeof(*flow->tid_entry) * flow->tidcnt;
	req->cur_seg++;

	memset(&ohdr->u.tid_rdma.w_rsp, 0, sizeof(ohdr->u.tid_rdma.w_rsp));
	epriv->ss.sge.vaddr = resp_addr;
	epriv->ss.sge.sge_length = resp_len;
	epriv->ss.sge.length = epriv->ss.sge.sge_length;
	/*
	 * We can safely zero these out. Since the first SGE covers the
	 * entire packet, nothing else should even look at the MR.
	 */
	epriv->ss.sge.mr = NULL;
	epriv->ss.sge.m = 0;
	epriv->ss.sge.n = 0;

	epriv->ss.sg_list = NULL;
	epriv->ss.total_len = epriv->ss.sge.sge_length;
	epriv->ss.num_sge = 1;

	*ss = &epriv->ss;
	*len = epriv->ss.total_len;

	/* Construct the TID RDMA WRITE RESP packet header */
	rcu_read_lock();
	remote = rcu_dereference(qpriv->tid_rdma.remote);

	KDETH_RESET(ohdr->u.tid_rdma.w_rsp.kdeth0, KVER, 0x1);
	KDETH_RESET(ohdr->u.tid_rdma.w_rsp.kdeth1, JKEY, remote->jkey);
	ohdr->u.tid_rdma.w_rsp.aeth = rvt_compute_aeth(qp);
	ohdr->u.tid_rdma.w_rsp.tid_flow_psn =
		cpu_to_be32((flow->flow_state.generation <<
			     HFI1_KDETH_BTH_SEQ_SHIFT) |
			    (flow->flow_state.spsn &
			     HFI1_KDETH_BTH_SEQ_MASK));
	ohdr->u.tid_rdma.w_rsp.tid_flow_qp =
		cpu_to_be32(qpriv->tid_rdma.local.qp |
			    ((flow->idx & TID_RDMA_DESTQP_FLOW_MASK) <<
			     TID_RDMA_DESTQP_FLOW_SHIFT) |
			    qpriv->rcd->ctxt);
	ohdr->u.tid_rdma.w_rsp.verbs_qp = cpu_to_be32(qp->remote_qpn);
	*bth1 = remote->qp;
	rcu_read_unlock();
	hdwords = sizeof(ohdr->u.tid_rdma.w_rsp) / sizeof(u32);
	qpriv->pending_tid_w_segs++;
done:
	return hdwords;
}

static void hfi1_add_tid_reap_timer(struct rvt_qp *qp)
{
	struct hfi1_qp_priv *qpriv = qp->priv;

	lockdep_assert_held(&qp->s_lock);
	if (!(qpriv->s_flags & HFI1_R_TID_RSC_TIMER)) {
		qpriv->s_flags |= HFI1_R_TID_RSC_TIMER;
		qpriv->s_tid_timer.expires = jiffies +
			qpriv->tid_timer_timeout_jiffies;
		add_timer(&qpriv->s_tid_timer);
	}
}

static void hfi1_mod_tid_reap_timer(struct rvt_qp *qp)
{
	struct hfi1_qp_priv *qpriv = qp->priv;

	lockdep_assert_held(&qp->s_lock);
	qpriv->s_flags |= HFI1_R_TID_RSC_TIMER;
	mod_timer(&qpriv->s_tid_timer, jiffies +
		  qpriv->tid_timer_timeout_jiffies);
}

static int hfi1_stop_tid_reap_timer(struct rvt_qp *qp)
{
	struct hfi1_qp_priv *qpriv = qp->priv;
	int rval = 0;

	lockdep_assert_held(&qp->s_lock);
	if (qpriv->s_flags & HFI1_R_TID_RSC_TIMER) {
		rval = del_timer(&qpriv->s_tid_timer);
		qpriv->s_flags &= ~HFI1_R_TID_RSC_TIMER;
	}
	return rval;
}

void hfi1_del_tid_reap_timer(struct rvt_qp *qp)
{
	struct hfi1_qp_priv *qpriv = qp->priv;

	del_timer_sync(&qpriv->s_tid_timer);
	qpriv->s_flags &= ~HFI1_R_TID_RSC_TIMER;
}

static void hfi1_tid_timeout(struct timer_list *t)
{
	struct hfi1_qp_priv *qpriv = from_timer(qpriv, t, s_tid_timer);
	struct rvt_qp *qp = qpriv->owner;
	struct rvt_dev_info *rdi = ib_to_rvt(qp->ibqp.device);
	unsigned long flags;
	u32 i;

	spin_lock_irqsave(&qp->r_lock, flags);
	spin_lock(&qp->s_lock);
	if (qpriv->s_flags & HFI1_R_TID_RSC_TIMER) {
		dd_dev_warn(dd_from_ibdev(qp->ibqp.device), "[QP%u] %s %d\n",
			    qp->ibqp.qp_num, __func__, __LINE__);
		trace_hfi1_msg_tid_timeout(/* msg */
			qp, "resource timeout = ",
			(u64)qpriv->tid_timer_timeout_jiffies);
		hfi1_stop_tid_reap_timer(qp);
		/*
		 * Go though the entire ack queue and clear any outstanding
		 * HW flow and RcvArray resources.
		 */
		hfi1_kern_clear_hw_flow(qpriv->rcd, qp);
		for (i = 0; i < rvt_max_atomic(rdi); i++) {
			struct tid_rdma_request *req =
				ack_to_tid_req(&qp->s_ack_queue[i]);

			hfi1_kern_exp_rcv_clear_all(req);
		}
		spin_unlock(&qp->s_lock);
		if (qp->ibqp.event_handler) {
			struct ib_event ev;

			ev.device = qp->ibqp.device;
			ev.element.qp = &qp->ibqp;
			ev.event = IB_EVENT_QP_FATAL;
			qp->ibqp.event_handler(&ev, qp->ibqp.qp_context);
		}
		rvt_rc_error(qp, IB_WC_RESP_TIMEOUT_ERR);
		goto unlock_r_lock;
	}
	spin_unlock(&qp->s_lock);
unlock_r_lock:
	spin_unlock_irqrestore(&qp->r_lock, flags);
}

void hfi1_rc_rcv_tid_rdma_write_resp(struct hfi1_packet *packet)
{
	/* HANDLER FOR TID RDMA WRITE RESPONSE packet (Requestor side */

	/*
	 * 1. Find matching SWQE
	 * 2. Check that TIDENTRY array has enough space for a complete
	 *    segment. If not, put QP in error state.
	 * 3. Save response data in struct tid_rdma_req and struct tid_rdma_flow
	 * 4. Remove HFI1_S_WAIT_TID_RESP from s_flags.
	 * 5. Set qp->s_state
	 * 6. Kick the send engine (hfi1_schedule_send())
	 */
	struct ib_other_headers *ohdr = packet->ohdr;
	struct rvt_qp *qp = packet->qp;
	struct hfi1_qp_priv *qpriv = qp->priv;
	struct hfi1_ctxtdata *rcd = packet->rcd;
	struct rvt_swqe *wqe;
	struct tid_rdma_request *req;
	struct tid_rdma_flow *flow;
	enum ib_wc_status status;
	u32 opcode, aeth, psn, flow_psn, i, tidlen = 0, pktlen;
	bool fecn;
	unsigned long flags;

	fecn = process_ecn(qp, packet);
	psn = mask_psn(be32_to_cpu(ohdr->bth[2]));
	aeth = be32_to_cpu(ohdr->u.tid_rdma.w_rsp.aeth);
	opcode = (be32_to_cpu(ohdr->bth[0]) >> 24) & 0xff;

	spin_lock_irqsave(&qp->s_lock, flags);

	/* Ignore invalid responses */
	if (cmp_psn(psn, qp->s_next_psn) >= 0)
		goto ack_done;

	/* Ignore duplicate responses. */
	if (unlikely(cmp_psn(psn, qp->s_last_psn) <= 0))
		goto ack_done;

	if (unlikely(qp->s_acked == qp->s_tail))
		goto ack_done;

	/*
	 * If we are waiting for a particular packet sequence number
	 * due to a request being resent, check for it. Otherwise,
	 * ensure that we haven't missed anything.
	 */
	if (qp->r_flags & RVT_R_RDMAR_SEQ) {
		if (cmp_psn(psn, qp->s_last_psn + 1) != 0)
			goto ack_done;
		qp->r_flags &= ~RVT_R_RDMAR_SEQ;
	}

	wqe = rvt_get_swqe_ptr(qp, qpriv->s_tid_cur);
	if (unlikely(wqe->wr.opcode != IB_WR_TID_RDMA_WRITE))
		goto ack_op_err;

	req = wqe_to_tid_req(wqe);
	/*
	 * If we've lost ACKs and our acked_tail pointer is too far
	 * behind, don't overwrite segments. Just drop the packet and
	 * let the reliability protocol take care of it.
	 */
	if (!CIRC_SPACE(req->setup_head, req->acked_tail, MAX_FLOWS))
		goto ack_done;

	/*
	 * The call to do_rc_ack() should be last in the chain of
	 * packet checks because it will end up updating the QP state.
	 * Therefore, anything that would prevent the packet from
	 * being accepted as a successful response should be prior
	 * to it.
	 */
	if (!do_rc_ack(qp, aeth, psn, opcode, 0, rcd))
		goto ack_done;

	trace_hfi1_ack(qp, psn);

	flow = &req->flows[req->setup_head];
	flow->pkt = 0;
	flow->tid_idx = 0;
	flow->tid_offset = 0;
	flow->sent = 0;
	flow->resync_npkts = 0;
	flow->tid_qpn = be32_to_cpu(ohdr->u.tid_rdma.w_rsp.tid_flow_qp);
	flow->idx = (flow->tid_qpn >> TID_RDMA_DESTQP_FLOW_SHIFT) &
		TID_RDMA_DESTQP_FLOW_MASK;
	flow_psn = mask_psn(be32_to_cpu(ohdr->u.tid_rdma.w_rsp.tid_flow_psn));
	flow->flow_state.generation = flow_psn >> HFI1_KDETH_BTH_SEQ_SHIFT;
	flow->flow_state.spsn = flow_psn & HFI1_KDETH_BTH_SEQ_MASK;
	flow->flow_state.resp_ib_psn = psn;
	flow->length = min_t(u32, req->seg_len,
			     (wqe->length - (req->comp_seg * req->seg_len)));

	flow->npkts = rvt_div_round_up_mtu(qp, flow->length);
	flow->flow_state.lpsn = flow->flow_state.spsn +
		flow->npkts - 1;
	/* payload length = packet length - (header length + ICRC length) */
	pktlen = packet->tlen - (packet->hlen + 4);
	if (pktlen > sizeof(flow->tid_entry)) {
		status = IB_WC_LOC_LEN_ERR;
		goto ack_err;
	}
	memcpy(flow->tid_entry, packet->ebuf, pktlen);
	flow->tidcnt = pktlen / sizeof(*flow->tid_entry);
	trace_hfi1_tid_flow_rcv_write_resp(qp, req->setup_head, flow);

	req->comp_seg++;
	trace_hfi1_tid_write_sender_rcv_resp(qp, 0);
	/*
	 * Walk the TID_ENTRY list to make sure we have enough space for a
	 * complete segment.
	 */
	for (i = 0; i < flow->tidcnt; i++) {
		trace_hfi1_tid_entry_rcv_write_resp(/* entry */
			qp, i, flow->tid_entry[i]);
		if (!EXP_TID_GET(flow->tid_entry[i], LEN)) {
			status = IB_WC_LOC_LEN_ERR;
			goto ack_err;
		}
		tidlen += EXP_TID_GET(flow->tid_entry[i], LEN);
	}
	if (tidlen * PAGE_SIZE < flow->length) {
		status = IB_WC_LOC_LEN_ERR;
		goto ack_err;
	}

	trace_hfi1_tid_req_rcv_write_resp(qp, 0, wqe->wr.opcode, wqe->psn,
					  wqe->lpsn, req);
	/*
	 * If this is the first response for this request, set the initial
	 * flow index to the current flow.
	 */
	if (!cmp_psn(psn, wqe->psn)) {
		req->r_last_acked = mask_psn(wqe->psn - 1);
		/* Set acked flow index to head index */
		req->acked_tail = req->setup_head;
	}

	/* advance circular buffer head */
	req->setup_head = CIRC_NEXT(req->setup_head, MAX_FLOWS);
	req->state = TID_REQUEST_ACTIVE;

	/*
	 * If all responses for this TID RDMA WRITE request have been received
	 * advance the pointer to the next one.
	 * Since TID RDMA requests could be mixed in with regular IB requests,
	 * they might not appear sequentially in the queue. Therefore, the
	 * next request needs to be "found".
	 */
	if (qpriv->s_tid_cur != qpriv->s_tid_head &&
	    req->comp_seg == req->total_segs) {
		for (i = qpriv->s_tid_cur + 1; ; i++) {
			if (i == qp->s_size)
				i = 0;
			wqe = rvt_get_swqe_ptr(qp, i);
			if (i == qpriv->s_tid_head)
				break;
			if (wqe->wr.opcode == IB_WR_TID_RDMA_WRITE)
				break;
		}
		qpriv->s_tid_cur = i;
	}
	qp->s_flags &= ~HFI1_S_WAIT_TID_RESP;
	hfi1_schedule_tid_send(qp);
	goto ack_done;

ack_op_err:
	status = IB_WC_LOC_QP_OP_ERR;
ack_err:
	rvt_error_qp(qp, status);
ack_done:
	if (fecn)
		qp->s_flags |= RVT_S_ECN;
	spin_unlock_irqrestore(&qp->s_lock, flags);
}

bool hfi1_build_tid_rdma_packet(struct rvt_swqe *wqe,
				struct ib_other_headers *ohdr,
				u32 *bth1, u32 *bth2, u32 *len)
{
	struct tid_rdma_request *req = wqe_to_tid_req(wqe);
	struct tid_rdma_flow *flow = &req->flows[req->clear_tail];
	struct tid_rdma_params *remote;
	struct rvt_qp *qp = req->qp;
	struct hfi1_qp_priv *qpriv = qp->priv;
	u32 tidentry = flow->tid_entry[flow->tid_idx];
	u32 tidlen = EXP_TID_GET(tidentry, LEN) << PAGE_SHIFT;
	struct tid_rdma_write_data *wd = &ohdr->u.tid_rdma.w_data;
	u32 next_offset, om = KDETH_OM_LARGE;
	bool last_pkt;

	if (!tidlen) {
		hfi1_trdma_send_complete(qp, wqe, IB_WC_REM_INV_RD_REQ_ERR);
		rvt_error_qp(qp, IB_WC_REM_INV_RD_REQ_ERR);
	}

	*len = min_t(u32, qp->pmtu, tidlen - flow->tid_offset);
	flow->sent += *len;
	next_offset = flow->tid_offset + *len;
	last_pkt = (flow->tid_idx == (flow->tidcnt - 1) &&
		    next_offset >= tidlen) || (flow->sent >= flow->length);
	trace_hfi1_tid_entry_build_write_data(qp, flow->tid_idx, tidentry);
	trace_hfi1_tid_flow_build_write_data(qp, req->clear_tail, flow);

	rcu_read_lock();
	remote = rcu_dereference(qpriv->tid_rdma.remote);
	KDETH_RESET(wd->kdeth0, KVER, 0x1);
	KDETH_SET(wd->kdeth0, SH, !last_pkt);
	KDETH_SET(wd->kdeth0, INTR, !!(!last_pkt && remote->urg));
	KDETH_SET(wd->kdeth0, TIDCTRL, EXP_TID_GET(tidentry, CTRL));
	KDETH_SET(wd->kdeth0, TID, EXP_TID_GET(tidentry, IDX));
	KDETH_SET(wd->kdeth0, OM, om == KDETH_OM_LARGE);
	KDETH_SET(wd->kdeth0, OFFSET, flow->tid_offset / om);
	KDETH_RESET(wd->kdeth1, JKEY, remote->jkey);
	wd->verbs_qp = cpu_to_be32(qp->remote_qpn);
	rcu_read_unlock();

	*bth1 = flow->tid_qpn;
	*bth2 = mask_psn(((flow->flow_state.spsn + flow->pkt++) &
			 HFI1_KDETH_BTH_SEQ_MASK) |
			 (flow->flow_state.generation <<
			  HFI1_KDETH_BTH_SEQ_SHIFT));
	if (last_pkt) {
		/* PSNs are zero-based, so +1 to count number of packets */
		if (flow->flow_state.lpsn + 1 +
		    rvt_div_round_up_mtu(qp, req->seg_len) >
		    MAX_TID_FLOW_PSN)
			req->state = TID_REQUEST_SYNC;
		*bth2 |= IB_BTH_REQ_ACK;
	}

	if (next_offset >= tidlen) {
		flow->tid_offset = 0;
		flow->tid_idx++;
	} else {
		flow->tid_offset = next_offset;
	}
	return last_pkt;
}

void hfi1_rc_rcv_tid_rdma_write_data(struct hfi1_packet *packet)
{
	struct rvt_qp *qp = packet->qp;
	struct hfi1_qp_priv *priv = qp->priv;
	struct hfi1_ctxtdata *rcd = priv->rcd;
	struct ib_other_headers *ohdr = packet->ohdr;
	struct rvt_ack_entry *e;
	struct tid_rdma_request *req;
	struct tid_rdma_flow *flow;
	struct hfi1_ibdev *dev = to_idev(qp->ibqp.device);
	unsigned long flags;
	u32 psn, next;
	u8 opcode;
	bool fecn;

	fecn = process_ecn(qp, packet);
	psn = mask_psn(be32_to_cpu(ohdr->bth[2]));
	opcode = (be32_to_cpu(ohdr->bth[0]) >> 24) & 0xff;

	/*
	 * All error handling should be done by now. If we are here, the packet
	 * is either good or been accepted by the error handler.
	 */
	spin_lock_irqsave(&qp->s_lock, flags);
	e = &qp->s_ack_queue[priv->r_tid_tail];
	req = ack_to_tid_req(e);
	flow = &req->flows[req->clear_tail];
	if (cmp_psn(psn, full_flow_psn(flow, flow->flow_state.lpsn))) {
		update_r_next_psn_fecn(packet, priv, rcd, flow, fecn);

		if (cmp_psn(psn, flow->flow_state.r_next_psn))
			goto send_nak;

		flow->flow_state.r_next_psn = mask_psn(psn + 1);
		/*
		 * Copy the payload to destination buffer if this packet is
		 * delivered as an eager packet due to RSM rule and FECN.
		 * The RSM rule selects FECN bit in BTH and SH bit in
		 * KDETH header and therefore will not match the last
		 * packet of each segment that has SH bit cleared.
		 */
		if (fecn && packet->etype == RHF_RCV_TYPE_EAGER) {
			struct rvt_sge_state ss;
			u32 len;
			u32 tlen = packet->tlen;
			u16 hdrsize = packet->hlen;
			u8 pad = packet->pad;
			u8 extra_bytes = pad + packet->extra_byte +
				(SIZE_OF_CRC << 2);
			u32 pmtu = qp->pmtu;

			if (unlikely(tlen != (hdrsize + pmtu + extra_bytes)))
				goto send_nak;
			len = req->comp_seg * req->seg_len;
			len += delta_psn(psn,
				full_flow_psn(flow, flow->flow_state.spsn)) *
				pmtu;
			if (unlikely(req->total_len - len < pmtu))
				goto send_nak;

			/*
			 * The e->rdma_sge field is set when TID RDMA WRITE REQ
			 * is first received and is never modified thereafter.
			 */
			ss.sge = e->rdma_sge;
			ss.sg_list = NULL;
			ss.num_sge = 1;
			ss.total_len = req->total_len;
			rvt_skip_sge(&ss, len, false);
			rvt_copy_sge(qp, &ss, packet->payload, pmtu, false,
				     false);
			/* Raise the sw sequence check flag for next packet */
			priv->r_next_psn_kdeth = mask_psn(psn + 1);
			priv->s_flags |= HFI1_R_TID_SW_PSN;
		}
		goto exit;
	}
	flow->flow_state.r_next_psn = mask_psn(psn + 1);
	hfi1_kern_exp_rcv_clear(req);
	priv->alloc_w_segs--;
	rcd->flows[flow->idx].psn = psn & HFI1_KDETH_BTH_SEQ_MASK;
	req->comp_seg++;
	priv->s_nak_state = 0;

	/*
	 * Release the flow if one of the following conditions has been met:
	 *  - The request has reached a sync point AND all outstanding
	 *    segments have been completed, or
	 *  - The entire request is complete and there are no more requests
	 *    (of any kind) in the queue.
	 */
	trace_hfi1_rsp_rcv_tid_write_data(qp, psn);
	trace_hfi1_tid_req_rcv_write_data(qp, 0, e->opcode, e->psn, e->lpsn,
					  req);
	trace_hfi1_tid_write_rsp_rcv_data(qp);
	if (priv->r_tid_ack == HFI1_QP_WQE_INVALID)
		priv->r_tid_ack = priv->r_tid_tail;

	if (opcode == TID_OP(WRITE_DATA_LAST)) {
		release_rdma_sge_mr(e);
		for (next = priv->r_tid_tail + 1; ; next++) {
			if (next > rvt_size_atomic(&dev->rdi))
				next = 0;
			if (next == priv->r_tid_head)
				break;
			e = &qp->s_ack_queue[next];
			if (e->opcode == TID_OP(WRITE_REQ))
				break;
		}
		priv->r_tid_tail = next;
		if (++qp->s_acked_ack_queue > rvt_size_atomic(&dev->rdi))
			qp->s_acked_ack_queue = 0;
	}

	hfi1_tid_write_alloc_resources(qp, true);

	/*
	 * If we need to generate more responses, schedule the
	 * send engine.
	 */
	if (req->cur_seg < req->total_segs ||
	    qp->s_tail_ack_queue != qp->r_head_ack_queue) {
		qp->s_flags |= RVT_S_RESP_PENDING;
		hfi1_schedule_send(qp);
	}

	priv->pending_tid_w_segs--;
	if (priv->s_flags & HFI1_R_TID_RSC_TIMER) {
		if (priv->pending_tid_w_segs)
			hfi1_mod_tid_reap_timer(req->qp);
		else
			hfi1_stop_tid_reap_timer(req->qp);
	}

done:
	priv->s_flags |= RVT_S_ACK_PENDING;
	hfi1_schedule_tid_send(qp);
exit:
	priv->r_next_psn_kdeth = flow->flow_state.r_next_psn;
	if (fecn)
		qp->s_flags |= RVT_S_ECN;
	spin_unlock_irqrestore(&qp->s_lock, flags);
	return;

send_nak:
	if (!priv->s_nak_state) {
		priv->s_nak_state = IB_NAK_PSN_ERROR;
		priv->s_nak_psn = flow->flow_state.r_next_psn;
		priv->s_flags |= RVT_S_ACK_PENDING;
		if (priv->r_tid_ack == HFI1_QP_WQE_INVALID)
			priv->r_tid_ack = priv->r_tid_tail;
		hfi1_schedule_tid_send(qp);
	}
	goto done;
}

static bool hfi1_tid_rdma_is_resync_psn(u32 psn)
{
	return (bool)((psn & HFI1_KDETH_BTH_SEQ_MASK) ==
		      HFI1_KDETH_BTH_SEQ_MASK);
}

u32 hfi1_build_tid_rdma_write_ack(struct rvt_qp *qp, struct rvt_ack_entry *e,
				  struct ib_other_headers *ohdr, u16 iflow,
				  u32 *bth1, u32 *bth2)
{
	struct hfi1_qp_priv *qpriv = qp->priv;
	struct tid_flow_state *fs = &qpriv->flow_state;
	struct tid_rdma_request *req = ack_to_tid_req(e);
	struct tid_rdma_flow *flow = &req->flows[iflow];
	struct tid_rdma_params *remote;

	rcu_read_lock();
	remote = rcu_dereference(qpriv->tid_rdma.remote);
	KDETH_RESET(ohdr->u.tid_rdma.ack.kdeth1, JKEY, remote->jkey);
	ohdr->u.tid_rdma.ack.verbs_qp = cpu_to_be32(qp->remote_qpn);
	*bth1 = remote->qp;
	rcu_read_unlock();

	if (qpriv->resync) {
		*bth2 = mask_psn((fs->generation <<
				  HFI1_KDETH_BTH_SEQ_SHIFT) - 1);
		ohdr->u.tid_rdma.ack.aeth = rvt_compute_aeth(qp);
	} else if (qpriv->s_nak_state) {
		*bth2 = mask_psn(qpriv->s_nak_psn);
		ohdr->u.tid_rdma.ack.aeth =
			cpu_to_be32((qp->r_msn & IB_MSN_MASK) |
				    (qpriv->s_nak_state <<
				     IB_AETH_CREDIT_SHIFT));
	} else {
		*bth2 = full_flow_psn(flow, flow->flow_state.lpsn);
		ohdr->u.tid_rdma.ack.aeth = rvt_compute_aeth(qp);
	}
	KDETH_RESET(ohdr->u.tid_rdma.ack.kdeth0, KVER, 0x1);
	ohdr->u.tid_rdma.ack.tid_flow_qp =
		cpu_to_be32(qpriv->tid_rdma.local.qp |
			    ((flow->idx & TID_RDMA_DESTQP_FLOW_MASK) <<
			     TID_RDMA_DESTQP_FLOW_SHIFT) |
			    qpriv->rcd->ctxt);

	ohdr->u.tid_rdma.ack.tid_flow_psn = 0;
	ohdr->u.tid_rdma.ack.verbs_psn =
		cpu_to_be32(flow->flow_state.resp_ib_psn);

	if (qpriv->resync) {
		/*
		 * If the PSN before the current expect KDETH PSN is the
		 * RESYNC PSN, then we never received a good TID RDMA WRITE
		 * DATA packet after a previous RESYNC.
		 * In this case, the next expected KDETH PSN stays the same.
		 */
		if (hfi1_tid_rdma_is_resync_psn(qpriv->r_next_psn_kdeth - 1)) {
			ohdr->u.tid_rdma.ack.tid_flow_psn =
				cpu_to_be32(qpriv->r_next_psn_kdeth_save);
		} else {
			/*
			 * Because the KDETH PSNs jump during a RESYNC, it's
			 * not possible to infer (or compute) the previous value
			 * of r_next_psn_kdeth in the case of back-to-back
			 * RESYNC packets. Therefore, we save it.
			 */
			qpriv->r_next_psn_kdeth_save =
				qpriv->r_next_psn_kdeth - 1;
			ohdr->u.tid_rdma.ack.tid_flow_psn =
				cpu_to_be32(qpriv->r_next_psn_kdeth_save);
			qpriv->r_next_psn_kdeth = mask_psn(*bth2 + 1);
		}
		qpriv->resync = false;
	}

	return sizeof(ohdr->u.tid_rdma.ack) / sizeof(u32);
}

void hfi1_rc_rcv_tid_rdma_ack(struct hfi1_packet *packet)
{
	struct ib_other_headers *ohdr = packet->ohdr;
	struct rvt_qp *qp = packet->qp;
	struct hfi1_qp_priv *qpriv = qp->priv;
	struct rvt_swqe *wqe;
	struct tid_rdma_request *req;
	struct tid_rdma_flow *flow;
	u32 aeth, psn, req_psn, ack_psn, flpsn, resync_psn, ack_kpsn;
	unsigned long flags;
	u16 fidx;

	trace_hfi1_tid_write_sender_rcv_tid_ack(qp, 0);
	process_ecn(qp, packet);
	psn = mask_psn(be32_to_cpu(ohdr->bth[2]));
	aeth = be32_to_cpu(ohdr->u.tid_rdma.ack.aeth);
	req_psn = mask_psn(be32_to_cpu(ohdr->u.tid_rdma.ack.verbs_psn));
	resync_psn = mask_psn(be32_to_cpu(ohdr->u.tid_rdma.ack.tid_flow_psn));

	spin_lock_irqsave(&qp->s_lock, flags);
	trace_hfi1_rcv_tid_ack(qp, aeth, psn, req_psn, resync_psn);

	/* If we are waiting for an ACK to RESYNC, drop any other packets */
	if ((qp->s_flags & HFI1_S_WAIT_HALT) &&
	    cmp_psn(psn, qpriv->s_resync_psn))
		goto ack_op_err;

	ack_psn = req_psn;
	if (hfi1_tid_rdma_is_resync_psn(psn))
		ack_kpsn = resync_psn;
	else
		ack_kpsn = psn;
	if (aeth >> 29) {
		ack_psn--;
		ack_kpsn--;
	}

	if (unlikely(qp->s_acked == qp->s_tail))
		goto ack_op_err;

	wqe = rvt_get_swqe_ptr(qp, qp->s_acked);

	if (wqe->wr.opcode != IB_WR_TID_RDMA_WRITE)
		goto ack_op_err;

	req = wqe_to_tid_req(wqe);
	trace_hfi1_tid_req_rcv_tid_ack(qp, 0, wqe->wr.opcode, wqe->psn,
				       wqe->lpsn, req);
	flow = &req->flows[req->acked_tail];
	trace_hfi1_tid_flow_rcv_tid_ack(qp, req->acked_tail, flow);

	/* Drop stale ACK/NAK */
	if (cmp_psn(psn, full_flow_psn(flow, flow->flow_state.spsn)) < 0 ||
	    cmp_psn(req_psn, flow->flow_state.resp_ib_psn) < 0)
		goto ack_op_err;

	while (cmp_psn(ack_kpsn,
		       full_flow_psn(flow, flow->flow_state.lpsn)) >= 0 &&
	       req->ack_seg < req->cur_seg) {
		req->ack_seg++;
		/* advance acked segment pointer */
		req->acked_tail = CIRC_NEXT(req->acked_tail, MAX_FLOWS);
		req->r_last_acked = flow->flow_state.resp_ib_psn;
		trace_hfi1_tid_req_rcv_tid_ack(qp, 0, wqe->wr.opcode, wqe->psn,
					       wqe->lpsn, req);
		if (req->ack_seg == req->total_segs) {
			req->state = TID_REQUEST_COMPLETE;
			wqe = do_rc_completion(qp, wqe,
					       to_iport(qp->ibqp.device,
							qp->port_num));
			trace_hfi1_sender_rcv_tid_ack(qp);
			atomic_dec(&qpriv->n_tid_requests);
			if (qp->s_acked == qp->s_tail)
				break;
			if (wqe->wr.opcode != IB_WR_TID_RDMA_WRITE)
				break;
			req = wqe_to_tid_req(wqe);
		}
		flow = &req->flows[req->acked_tail];
		trace_hfi1_tid_flow_rcv_tid_ack(qp, req->acked_tail, flow);
	}

	trace_hfi1_tid_req_rcv_tid_ack(qp, 0, wqe->wr.opcode, wqe->psn,
				       wqe->lpsn, req);
	switch (aeth >> 29) {
	case 0:         /* ACK */
		if (qpriv->s_flags & RVT_S_WAIT_ACK)
			qpriv->s_flags &= ~RVT_S_WAIT_ACK;
		if (!hfi1_tid_rdma_is_resync_psn(psn)) {
			/* Check if there is any pending TID ACK */
			if (wqe->wr.opcode == IB_WR_TID_RDMA_WRITE &&
			    req->ack_seg < req->cur_seg)
				hfi1_mod_tid_retry_timer(qp);
			else
				hfi1_stop_tid_retry_timer(qp);
			hfi1_schedule_send(qp);
		} else {
			u32 spsn, fpsn, last_acked, generation;
			struct tid_rdma_request *rptr;

			/* ACK(RESYNC) */
			hfi1_stop_tid_retry_timer(qp);
			/* Allow new requests (see hfi1_make_tid_rdma_pkt) */
			qp->s_flags &= ~HFI1_S_WAIT_HALT;
			/*
			 * Clear RVT_S_SEND_ONE flag in case that the TID RDMA
			 * ACK is received after the TID retry timer is fired
			 * again. In this case, do not send any more TID
			 * RESYNC request or wait for any more TID ACK packet.
			 */
			qpriv->s_flags &= ~RVT_S_SEND_ONE;
			hfi1_schedule_send(qp);

			if ((qp->s_acked == qpriv->s_tid_tail &&
			     req->ack_seg == req->total_segs) ||
			    qp->s_acked == qp->s_tail) {
				qpriv->s_state = TID_OP(WRITE_DATA_LAST);
				goto done;
			}

			if (req->ack_seg == req->comp_seg) {
				qpriv->s_state = TID_OP(WRITE_DATA);
				goto done;
			}

			/*
			 * The PSN to start with is the next PSN after the
			 * RESYNC PSN.
			 */
			psn = mask_psn(psn + 1);
			generation = psn >> HFI1_KDETH_BTH_SEQ_SHIFT;
			spsn = 0;

			/*
			 * Update to the correct WQE when we get an ACK(RESYNC)
			 * in the middle of a request.
			 */
			if (delta_psn(ack_psn, wqe->lpsn))
				wqe = rvt_get_swqe_ptr(qp, qp->s_acked);
			req = wqe_to_tid_req(wqe);
			flow = &req->flows[req->acked_tail];
			/*
			 * RESYNC re-numbers the PSN ranges of all remaining
			 * segments. Also, PSN's start from 0 in the middle of a
			 * segment and the first segment size is less than the
			 * default number of packets. flow->resync_npkts is used
			 * to track the number of packets from the start of the
			 * real segment to the point of 0 PSN after the RESYNC
			 * in order to later correctly rewind the SGE.
			 */
			fpsn = full_flow_psn(flow, flow->flow_state.spsn);
			req->r_ack_psn = psn;
			flow->resync_npkts +=
				delta_psn(mask_psn(resync_psn + 1), fpsn);
			/*
			 * Renumber all packet sequence number ranges
			 * based on the new generation.
			 */
			last_acked = qp->s_acked;
			rptr = req;
			while (1) {
				/* start from last acked segment */
				for (fidx = rptr->acked_tail;
				     CIRC_CNT(rptr->setup_head, fidx,
					      MAX_FLOWS);
				     fidx = CIRC_NEXT(fidx, MAX_FLOWS)) {
					u32 lpsn;
					u32 gen;

					flow = &rptr->flows[fidx];
					gen = flow->flow_state.generation;
					if (WARN_ON(gen == generation &&
						    flow->flow_state.spsn !=
						     spsn))
						continue;
					lpsn = flow->flow_state.lpsn;
					lpsn = full_flow_psn(flow, lpsn);
					flow->npkts =
						delta_psn(lpsn,
							  mask_psn(resync_psn)
							  );
					flow->flow_state.generation =
						generation;
					flow->flow_state.spsn = spsn;
					flow->flow_state.lpsn =
						flow->flow_state.spsn +
						flow->npkts - 1;
					flow->pkt = 0;
					spsn += flow->npkts;
					resync_psn += flow->npkts;
					trace_hfi1_tid_flow_rcv_tid_ack(qp,
									fidx,
									flow);
				}
				if (++last_acked == qpriv->s_tid_cur + 1)
					break;
				if (last_acked == qp->s_size)
					last_acked = 0;
				wqe = rvt_get_swqe_ptr(qp, last_acked);
				rptr = wqe_to_tid_req(wqe);
			}
			req->cur_seg = req->ack_seg;
			qpriv->s_tid_tail = qp->s_acked;
			qpriv->s_state = TID_OP(WRITE_REQ);
			hfi1_schedule_tid_send(qp);
		}
done:
		qpriv->s_retry = qp->s_retry_cnt;
		break;

	case 3:         /* NAK */
		hfi1_stop_tid_retry_timer(qp);
		switch ((aeth >> IB_AETH_CREDIT_SHIFT) &
			IB_AETH_CREDIT_MASK) {
		case 0: /* PSN sequence error */
			if (!req->flows)
				break;
			flow = &req->flows[req->acked_tail];
			flpsn = full_flow_psn(flow, flow->flow_state.lpsn);
			if (cmp_psn(psn, flpsn) > 0)
				break;
			trace_hfi1_tid_flow_rcv_tid_ack(qp, req->acked_tail,
							flow);
			req->r_ack_psn = mask_psn(be32_to_cpu(ohdr->bth[2]));
			req->cur_seg = req->ack_seg;
			qpriv->s_tid_tail = qp->s_acked;
			qpriv->s_state = TID_OP(WRITE_REQ);
			qpriv->s_retry = qp->s_retry_cnt;
			hfi1_schedule_tid_send(qp);
			break;

		default:
			break;
		}
		break;

	default:
		break;
	}

ack_op_err:
	spin_unlock_irqrestore(&qp->s_lock, flags);
}

void hfi1_add_tid_retry_timer(struct rvt_qp *qp)
{
	struct hfi1_qp_priv *priv = qp->priv;
	struct ib_qp *ibqp = &qp->ibqp;
	struct rvt_dev_info *rdi = ib_to_rvt(ibqp->device);

	lockdep_assert_held(&qp->s_lock);
	if (!(priv->s_flags & HFI1_S_TID_RETRY_TIMER)) {
		priv->s_flags |= HFI1_S_TID_RETRY_TIMER;
		priv->s_tid_retry_timer.expires = jiffies +
			priv->tid_retry_timeout_jiffies + rdi->busy_jiffies;
		add_timer(&priv->s_tid_retry_timer);
	}
}

static void hfi1_mod_tid_retry_timer(struct rvt_qp *qp)
{
	struct hfi1_qp_priv *priv = qp->priv;
	struct ib_qp *ibqp = &qp->ibqp;
	struct rvt_dev_info *rdi = ib_to_rvt(ibqp->device);

	lockdep_assert_held(&qp->s_lock);
	priv->s_flags |= HFI1_S_TID_RETRY_TIMER;
	mod_timer(&priv->s_tid_retry_timer, jiffies +
		  priv->tid_retry_timeout_jiffies + rdi->busy_jiffies);
}

static int hfi1_stop_tid_retry_timer(struct rvt_qp *qp)
{
	struct hfi1_qp_priv *priv = qp->priv;
	int rval = 0;

	lockdep_assert_held(&qp->s_lock);
	if (priv->s_flags & HFI1_S_TID_RETRY_TIMER) {
		rval = del_timer(&priv->s_tid_retry_timer);
		priv->s_flags &= ~HFI1_S_TID_RETRY_TIMER;
	}
	return rval;
}

void hfi1_del_tid_retry_timer(struct rvt_qp *qp)
{
	struct hfi1_qp_priv *priv = qp->priv;

	del_timer_sync(&priv->s_tid_retry_timer);
	priv->s_flags &= ~HFI1_S_TID_RETRY_TIMER;
}

static void hfi1_tid_retry_timeout(struct timer_list *t)
{
	struct hfi1_qp_priv *priv = from_timer(priv, t, s_tid_retry_timer);
	struct rvt_qp *qp = priv->owner;
	struct rvt_swqe *wqe;
	unsigned long flags;
	struct tid_rdma_request *req;

	spin_lock_irqsave(&qp->r_lock, flags);
	spin_lock(&qp->s_lock);
	trace_hfi1_tid_write_sender_retry_timeout(qp, 0);
	if (priv->s_flags & HFI1_S_TID_RETRY_TIMER) {
		hfi1_stop_tid_retry_timer(qp);
		if (!priv->s_retry) {
			trace_hfi1_msg_tid_retry_timeout(/* msg */
				qp,
				"Exhausted retries. Tid retry timeout = ",
				(u64)priv->tid_retry_timeout_jiffies);

			wqe = rvt_get_swqe_ptr(qp, qp->s_acked);
			hfi1_trdma_send_complete(qp, wqe, IB_WC_RETRY_EXC_ERR);
			rvt_error_qp(qp, IB_WC_WR_FLUSH_ERR);
		} else {
			wqe = rvt_get_swqe_ptr(qp, qp->s_acked);
			req = wqe_to_tid_req(wqe);
			trace_hfi1_tid_req_tid_retry_timeout(/* req */
			   qp, 0, wqe->wr.opcode, wqe->psn, wqe->lpsn, req);

			priv->s_flags &= ~RVT_S_WAIT_ACK;
			/* Only send one packet (the RESYNC) */
			priv->s_flags |= RVT_S_SEND_ONE;
			/*
			 * No additional request shall be made by this QP until
			 * the RESYNC has been complete.
			 */
			qp->s_flags |= HFI1_S_WAIT_HALT;
			priv->s_state = TID_OP(RESYNC);
			priv->s_retry--;
			hfi1_schedule_tid_send(qp);
		}
	}
	spin_unlock(&qp->s_lock);
	spin_unlock_irqrestore(&qp->r_lock, flags);
}

u32 hfi1_build_tid_rdma_resync(struct rvt_qp *qp, struct rvt_swqe *wqe,
			       struct ib_other_headers *ohdr, u32 *bth1,
			       u32 *bth2, u16 fidx)
{
	struct hfi1_qp_priv *qpriv = qp->priv;
	struct tid_rdma_params *remote;
	struct tid_rdma_request *req = wqe_to_tid_req(wqe);
	struct tid_rdma_flow *flow = &req->flows[fidx];
	u32 generation;

	rcu_read_lock();
	remote = rcu_dereference(qpriv->tid_rdma.remote);
	KDETH_RESET(ohdr->u.tid_rdma.ack.kdeth1, JKEY, remote->jkey);
	ohdr->u.tid_rdma.ack.verbs_qp = cpu_to_be32(qp->remote_qpn);
	*bth1 = remote->qp;
	rcu_read_unlock();

	generation = kern_flow_generation_next(flow->flow_state.generation);
	*bth2 = mask_psn((generation << HFI1_KDETH_BTH_SEQ_SHIFT) - 1);
	qpriv->s_resync_psn = *bth2;
	*bth2 |= IB_BTH_REQ_ACK;
	KDETH_RESET(ohdr->u.tid_rdma.ack.kdeth0, KVER, 0x1);

	return sizeof(ohdr->u.tid_rdma.resync) / sizeof(u32);
}

void hfi1_rc_rcv_tid_rdma_resync(struct hfi1_packet *packet)
{
	struct ib_other_headers *ohdr = packet->ohdr;
	struct rvt_qp *qp = packet->qp;
	struct hfi1_qp_priv *qpriv = qp->priv;
	struct hfi1_ctxtdata *rcd = qpriv->rcd;
	struct hfi1_ibdev *dev = to_idev(qp->ibqp.device);
	struct rvt_ack_entry *e;
	struct tid_rdma_request *req;
	struct tid_rdma_flow *flow;
	struct tid_flow_state *fs = &qpriv->flow_state;
	u32 psn, generation, idx, gen_next;
	bool fecn;
	unsigned long flags;

	fecn = process_ecn(qp, packet);
	psn = mask_psn(be32_to_cpu(ohdr->bth[2]));

	generation = mask_psn(psn + 1) >> HFI1_KDETH_BTH_SEQ_SHIFT;
	spin_lock_irqsave(&qp->s_lock, flags);

	gen_next = (fs->generation == KERN_GENERATION_RESERVED) ?
		generation : kern_flow_generation_next(fs->generation);
	/*
	 * RESYNC packet contains the "next" generation and can only be
	 * from the current or previous generations
	 */
	if (generation != mask_generation(gen_next - 1) &&
	    generation != gen_next)
		goto bail;
	/* Already processing a resync */
	if (qpriv->resync)
		goto bail;

	spin_lock(&rcd->exp_lock);
	if (fs->index >= RXE_NUM_TID_FLOWS) {
		/*
		 * If we don't have a flow, save the generation so it can be
		 * applied when a new flow is allocated
		 */
		fs->generation = generation;
	} else {
		/* Reprogram the QP flow with new generation */
		rcd->flows[fs->index].generation = generation;
		fs->generation = kern_setup_hw_flow(rcd, fs->index);
	}
	fs->psn = 0;
	/*
	 * Disable SW PSN checking since a RESYNC is equivalent to a
	 * sync point and the flow has/will be reprogrammed
	 */
	qpriv->s_flags &= ~HFI1_R_TID_SW_PSN;
	trace_hfi1_tid_write_rsp_rcv_resync(qp);

	/*
	 * Reset all TID flow information with the new generation.
	 * This is done for all requests and segments after the
	 * last received segment
	 */
	for (idx = qpriv->r_tid_tail; ; idx++) {
		u16 flow_idx;

		if (idx > rvt_size_atomic(&dev->rdi))
			idx = 0;
		e = &qp->s_ack_queue[idx];
		if (e->opcode == TID_OP(WRITE_REQ)) {
			req = ack_to_tid_req(e);
			trace_hfi1_tid_req_rcv_resync(qp, 0, e->opcode, e->psn,
						      e->lpsn, req);

			/* start from last unacked segment */
			for (flow_idx = req->clear_tail;
			     CIRC_CNT(req->setup_head, flow_idx,
				      MAX_FLOWS);
			     flow_idx = CIRC_NEXT(flow_idx, MAX_FLOWS)) {
				u32 lpsn;
				u32 next;

				flow = &req->flows[flow_idx];
				lpsn = full_flow_psn(flow,
						     flow->flow_state.lpsn);
				next = flow->flow_state.r_next_psn;
				flow->npkts = delta_psn(lpsn, next - 1);
				flow->flow_state.generation = fs->generation;
				flow->flow_state.spsn = fs->psn;
				flow->flow_state.lpsn =
					flow->flow_state.spsn + flow->npkts - 1;
				flow->flow_state.r_next_psn =
					full_flow_psn(flow,
						      flow->flow_state.spsn);
				fs->psn += flow->npkts;
				trace_hfi1_tid_flow_rcv_resync(qp, flow_idx,
							       flow);
			}
		}
		if (idx == qp->s_tail_ack_queue)
			break;
	}

	spin_unlock(&rcd->exp_lock);
	qpriv->resync = true;
	/* RESYNC request always gets a TID RDMA ACK. */
	qpriv->s_nak_state = 0;
	qpriv->s_flags |= RVT_S_ACK_PENDING;
	hfi1_schedule_tid_send(qp);
bail:
	if (fecn)
		qp->s_flags |= RVT_S_ECN;
	spin_unlock_irqrestore(&qp->s_lock, flags);
}

/*
 * Call this function when the last TID RDMA WRITE DATA packet for a request
 * is built.
 */
static void update_tid_tail(struct rvt_qp *qp)
	__must_hold(&qp->s_lock)
{
	struct hfi1_qp_priv *priv = qp->priv;
	u32 i;
	struct rvt_swqe *wqe;

	lockdep_assert_held(&qp->s_lock);
	/* Can't move beyond s_tid_cur */
	if (priv->s_tid_tail == priv->s_tid_cur)
		return;
	for (i = priv->s_tid_tail + 1; ; i++) {
		if (i == qp->s_size)
			i = 0;

		if (i == priv->s_tid_cur)
			break;
		wqe = rvt_get_swqe_ptr(qp, i);
		if (wqe->wr.opcode == IB_WR_TID_RDMA_WRITE)
			break;
	}
	priv->s_tid_tail = i;
	priv->s_state = TID_OP(WRITE_RESP);
}

int hfi1_make_tid_rdma_pkt(struct rvt_qp *qp, struct hfi1_pkt_state *ps)
	__must_hold(&qp->s_lock)
{
	struct hfi1_qp_priv *priv = qp->priv;
	struct rvt_swqe *wqe;
	u32 bth1 = 0, bth2 = 0, hwords = 5, len, middle = 0;
	struct ib_other_headers *ohdr;
	struct rvt_sge_state *ss = &qp->s_sge;
	struct rvt_ack_entry *e = &qp->s_ack_queue[qp->s_tail_ack_queue];
	struct tid_rdma_request *req = ack_to_tid_req(e);
	bool last = false;
	u8 opcode = TID_OP(WRITE_DATA);

	lockdep_assert_held(&qp->s_lock);
	trace_hfi1_tid_write_sender_make_tid_pkt(qp, 0);
	/*
	 * Prioritize the sending of the requests and responses over the
	 * sending of the TID RDMA data packets.
	 */
	if (((atomic_read(&priv->n_tid_requests) < HFI1_TID_RDMA_WRITE_CNT) &&
	     atomic_read(&priv->n_requests) &&
	     !(qp->s_flags & (RVT_S_BUSY | RVT_S_WAIT_ACK |
			     HFI1_S_ANY_WAIT_IO))) ||
	    (e->opcode == TID_OP(WRITE_REQ) && req->cur_seg < req->alloc_seg &&
	     !(qp->s_flags & (RVT_S_BUSY | HFI1_S_ANY_WAIT_IO)))) {
		struct iowait_work *iowork;

		iowork = iowait_get_ib_work(&priv->s_iowait);
		ps->s_txreq = get_waiting_verbs_txreq(iowork);
		if (ps->s_txreq || hfi1_make_rc_req(qp, ps)) {
			priv->s_flags |= HFI1_S_TID_BUSY_SET;
			return 1;
		}
	}

	ps->s_txreq = get_txreq(ps->dev, qp);
	if (!ps->s_txreq)
		goto bail_no_tx;

	ohdr = &ps->s_txreq->phdr.hdr.ibh.u.oth;

	if ((priv->s_flags & RVT_S_ACK_PENDING) &&
	    make_tid_rdma_ack(qp, ohdr, ps))
		return 1;

	/*
	 * Bail out if we can't send data.
	 * Be reminded that this check must been done after the call to
	 * make_tid_rdma_ack() because the responding QP could be in
	 * RTR state where it can send TID RDMA ACK, not TID RDMA WRITE DATA.
	 */
	if (!(ib_rvt_state_ops[qp->state] & RVT_PROCESS_SEND_OK))
		goto bail;

	if (priv->s_flags & RVT_S_WAIT_ACK)
		goto bail;

	/* Check whether there is anything to do. */
	if (priv->s_tid_tail == HFI1_QP_WQE_INVALID)
		goto bail;
	wqe = rvt_get_swqe_ptr(qp, priv->s_tid_tail);
	req = wqe_to_tid_req(wqe);
	trace_hfi1_tid_req_make_tid_pkt(qp, 0, wqe->wr.opcode, wqe->psn,
					wqe->lpsn, req);
	switch (priv->s_state) {
	case TID_OP(WRITE_REQ):
	case TID_OP(WRITE_RESP):
		priv->tid_ss.sge = wqe->sg_list[0];
		priv->tid_ss.sg_list = wqe->sg_list + 1;
		priv->tid_ss.num_sge = wqe->wr.num_sge;
		priv->tid_ss.total_len = wqe->length;

		if (priv->s_state == TID_OP(WRITE_REQ))
			hfi1_tid_rdma_restart_req(qp, wqe, &bth2);
		priv->s_state = TID_OP(WRITE_DATA);
		/* fall through */

	case TID_OP(WRITE_DATA):
		/*
		 * 1. Check whether TID RDMA WRITE RESP available.
		 * 2. If no:
		 *    2.1 If have more segments and no TID RDMA WRITE RESP,
		 *        set HFI1_S_WAIT_TID_RESP
		 *    2.2 Return indicating no progress made.
		 * 3. If yes:
		 *    3.1 Build TID RDMA WRITE DATA packet.
		 *    3.2 If last packet in segment:
		 *        3.2.1 Change KDETH header bits
		 *        3.2.2 Advance RESP pointers.
		 *    3.3 Return indicating progress made.
		 */
		trace_hfi1_sender_make_tid_pkt(qp);
		trace_hfi1_tid_write_sender_make_tid_pkt(qp, 0);
		wqe = rvt_get_swqe_ptr(qp, priv->s_tid_tail);
		req = wqe_to_tid_req(wqe);
		len = wqe->length;

		if (!req->comp_seg || req->cur_seg == req->comp_seg)
			goto bail;

		trace_hfi1_tid_req_make_tid_pkt(qp, 0, wqe->wr.opcode,
						wqe->psn, wqe->lpsn, req);
		last = hfi1_build_tid_rdma_packet(wqe, ohdr, &bth1, &bth2,
						  &len);

		if (last) {
			/* move pointer to next flow */
			req->clear_tail = CIRC_NEXT(req->clear_tail,
						    MAX_FLOWS);
			if (++req->cur_seg < req->total_segs) {
				if (!CIRC_CNT(req->setup_head, req->clear_tail,
					      MAX_FLOWS))
					qp->s_flags |= HFI1_S_WAIT_TID_RESP;
			} else {
				priv->s_state = TID_OP(WRITE_DATA_LAST);
				opcode = TID_OP(WRITE_DATA_LAST);

				/* Advance the s_tid_tail now */
				update_tid_tail(qp);
			}
		}
		hwords += sizeof(ohdr->u.tid_rdma.w_data) / sizeof(u32);
		ss = &priv->tid_ss;
		break;

	case TID_OP(RESYNC):
		trace_hfi1_sender_make_tid_pkt(qp);
		/* Use generation from the most recently received response */
		wqe = rvt_get_swqe_ptr(qp, priv->s_tid_cur);
		req = wqe_to_tid_req(wqe);
		/* If no responses for this WQE look at the previous one */
		if (!req->comp_seg) {
			wqe = rvt_get_swqe_ptr(qp,
					       (!priv->s_tid_cur ? qp->s_size :
						priv->s_tid_cur) - 1);
			req = wqe_to_tid_req(wqe);
		}
		hwords += hfi1_build_tid_rdma_resync(qp, wqe, ohdr, &bth1,
						     &bth2,
						     CIRC_PREV(req->setup_head,
							       MAX_FLOWS));
		ss = NULL;
		len = 0;
		opcode = TID_OP(RESYNC);
		break;

	default:
		goto bail;
	}
	if (priv->s_flags & RVT_S_SEND_ONE) {
		priv->s_flags &= ~RVT_S_SEND_ONE;
		priv->s_flags |= RVT_S_WAIT_ACK;
		bth2 |= IB_BTH_REQ_ACK;
	}
	qp->s_len -= len;
	ps->s_txreq->hdr_dwords = hwords;
	ps->s_txreq->sde = priv->s_sde;
	ps->s_txreq->ss = ss;
	ps->s_txreq->s_cur_size = len;
	hfi1_make_ruc_header(qp, ohdr, (opcode << 24), bth1, bth2,
			     middle, ps);
	return 1;
bail:
	hfi1_put_txreq(ps->s_txreq);
bail_no_tx:
	ps->s_txreq = NULL;
	priv->s_flags &= ~RVT_S_BUSY;
	/*
	 * If we didn't get a txreq, the QP will be woken up later to try
	 * again, set the flags to the the wake up which work item to wake
	 * up.
	 * (A better algorithm should be found to do this and generalize the
	 * sleep/wakeup flags.)
	 */
	iowait_set_flag(&priv->s_iowait, IOWAIT_PENDING_TID);
	return 0;
}

static int make_tid_rdma_ack(struct rvt_qp *qp,
			     struct ib_other_headers *ohdr,
			     struct hfi1_pkt_state *ps)
{
	struct rvt_ack_entry *e;
	struct hfi1_qp_priv *qpriv = qp->priv;
	struct hfi1_ibdev *dev = to_idev(qp->ibqp.device);
	u32 hwords, next;
	u32 len = 0;
	u32 bth1 = 0, bth2 = 0;
	int middle = 0;
	u16 flow;
	struct tid_rdma_request *req, *nreq;

	trace_hfi1_tid_write_rsp_make_tid_ack(qp);
	/* Don't send an ACK if we aren't supposed to. */
	if (!(ib_rvt_state_ops[qp->state] & RVT_PROCESS_RECV_OK))
		goto bail;

	/* header size in 32-bit words LRH+BTH = (8+12)/4. */
	hwords = 5;

	e = &qp->s_ack_queue[qpriv->r_tid_ack];
	req = ack_to_tid_req(e);
	/*
	 * In the RESYNC case, we are exactly one segment past the
	 * previously sent ack or at the previously sent NAK. So to send
	 * the resync ack, we go back one segment (which might be part of
	 * the previous request) and let the do-while loop execute again.
	 * The advantage of executing the do-while loop is that any data
	 * received after the previous ack is automatically acked in the
	 * RESYNC ack. It turns out that for the do-while loop we only need
	 * to pull back qpriv->r_tid_ack, not the segment
	 * indices/counters. The scheme works even if the previous request
	 * was not a TID WRITE request.
	 */
	if (qpriv->resync) {
		if (!req->ack_seg || req->ack_seg == req->total_segs)
			qpriv->r_tid_ack = !qpriv->r_tid_ack ?
				rvt_size_atomic(&dev->rdi) :
				qpriv->r_tid_ack - 1;
		e = &qp->s_ack_queue[qpriv->r_tid_ack];
		req = ack_to_tid_req(e);
	}

	trace_hfi1_rsp_make_tid_ack(qp, e->psn);
	trace_hfi1_tid_req_make_tid_ack(qp, 0, e->opcode, e->psn, e->lpsn,
					req);
	/*
	 * If we've sent all the ACKs that we can, we are done
	 * until we get more segments...
	 */
	if (!qpriv->s_nak_state && !qpriv->resync &&
	    req->ack_seg == req->comp_seg)
		goto bail;

	do {
		/*
		 * To deal with coalesced ACKs, the acked_tail pointer
		 * into the flow array is used. The distance between it
		 * and the clear_tail is the number of flows that are
		 * being ACK'ed.
		 */
		req->ack_seg +=
			/* Get up-to-date value */
			CIRC_CNT(req->clear_tail, req->acked_tail,
				 MAX_FLOWS);
		/* Advance acked index */
		req->acked_tail = req->clear_tail;

		/*
		 * req->clear_tail points to the segment currently being
		 * received. So, when sending an ACK, the previous
		 * segment is being ACK'ed.
		 */
		flow = CIRC_PREV(req->acked_tail, MAX_FLOWS);
		if (req->ack_seg != req->total_segs)
			break;
		req->state = TID_REQUEST_COMPLETE;

		next = qpriv->r_tid_ack + 1;
		if (next > rvt_size_atomic(&dev->rdi))
			next = 0;
		qpriv->r_tid_ack = next;
		if (qp->s_ack_queue[next].opcode != TID_OP(WRITE_REQ))
			break;
		nreq = ack_to_tid_req(&qp->s_ack_queue[next]);
		if (!nreq->comp_seg || nreq->ack_seg == nreq->comp_seg)
			break;

		/* Move to the next ack entry now */
		e = &qp->s_ack_queue[qpriv->r_tid_ack];
		req = ack_to_tid_req(e);
	} while (1);

	/*
	 * At this point qpriv->r_tid_ack == qpriv->r_tid_tail but e and
	 * req could be pointing at the previous ack queue entry
	 */
	if (qpriv->s_nak_state ||
	    (qpriv->resync &&
	     !hfi1_tid_rdma_is_resync_psn(qpriv->r_next_psn_kdeth - 1) &&
	     (cmp_psn(qpriv->r_next_psn_kdeth - 1,
		      full_flow_psn(&req->flows[flow],
				    req->flows[flow].flow_state.lpsn)) > 0))) {
		/*
		 * A NAK will implicitly acknowledge all previous TID RDMA
		 * requests. Therefore, we NAK with the req->acked_tail
		 * segment for the request at qpriv->r_tid_ack (same at
		 * this point as the req->clear_tail segment for the
		 * qpriv->r_tid_tail request)
		 */
		e = &qp->s_ack_queue[qpriv->r_tid_ack];
		req = ack_to_tid_req(e);
		flow = req->acked_tail;
	} else if (req->ack_seg == req->total_segs &&
		   qpriv->s_flags & HFI1_R_TID_WAIT_INTERLCK)
		qpriv->s_flags &= ~HFI1_R_TID_WAIT_INTERLCK;

	trace_hfi1_tid_write_rsp_make_tid_ack(qp);
	trace_hfi1_tid_req_make_tid_ack(qp, 0, e->opcode, e->psn, e->lpsn,
					req);
	hwords += hfi1_build_tid_rdma_write_ack(qp, e, ohdr, flow, &bth1,
						&bth2);
	len = 0;
	qpriv->s_flags &= ~RVT_S_ACK_PENDING;
	ps->s_txreq->hdr_dwords = hwords;
	ps->s_txreq->sde = qpriv->s_sde;
	ps->s_txreq->s_cur_size = len;
	ps->s_txreq->ss = NULL;
	hfi1_make_ruc_header(qp, ohdr, (TID_OP(ACK) << 24), bth1, bth2, middle,
			     ps);
	ps->s_txreq->txreq.flags |= SDMA_TXREQ_F_VIP;
	return 1;
bail:
	/*
	 * Ensure s_rdma_ack_cnt changes are committed prior to resetting
	 * RVT_S_RESP_PENDING
	 */
	smp_wmb();
	qpriv->s_flags &= ~RVT_S_ACK_PENDING;
	return 0;
}

static int hfi1_send_tid_ok(struct rvt_qp *qp)
{
	struct hfi1_qp_priv *priv = qp->priv;

	return !(priv->s_flags & RVT_S_BUSY ||
		 qp->s_flags & HFI1_S_ANY_WAIT_IO) &&
		(verbs_txreq_queued(iowait_get_tid_work(&priv->s_iowait)) ||
		 (priv->s_flags & RVT_S_RESP_PENDING) ||
		 !(qp->s_flags & HFI1_S_ANY_TID_WAIT_SEND));
}

void _hfi1_do_tid_send(struct work_struct *work)
{
	struct iowait_work *w = container_of(work, struct iowait_work, iowork);
	struct rvt_qp *qp = iowait_to_qp(w->iow);

	hfi1_do_tid_send(qp);
}

static void hfi1_do_tid_send(struct rvt_qp *qp)
{
	struct hfi1_pkt_state ps;
	struct hfi1_qp_priv *priv = qp->priv;

	ps.dev = to_idev(qp->ibqp.device);
	ps.ibp = to_iport(qp->ibqp.device, qp->port_num);
	ps.ppd = ppd_from_ibp(ps.ibp);
	ps.wait = iowait_get_tid_work(&priv->s_iowait);
	ps.in_thread = false;
	ps.timeout_int = qp->timeout_jiffies / 8;

	trace_hfi1_rc_do_tid_send(qp, false);
	spin_lock_irqsave(&qp->s_lock, ps.flags);

	/* Return if we are already busy processing a work request. */
	if (!hfi1_send_tid_ok(qp)) {
		if (qp->s_flags & HFI1_S_ANY_WAIT_IO)
			iowait_set_flag(&priv->s_iowait, IOWAIT_PENDING_TID);
		spin_unlock_irqrestore(&qp->s_lock, ps.flags);
		return;
	}

	priv->s_flags |= RVT_S_BUSY;

	ps.timeout = jiffies + ps.timeout_int;
	ps.cpu = priv->s_sde ? priv->s_sde->cpu :
		cpumask_first(cpumask_of_node(ps.ppd->dd->node));
	ps.pkts_sent = false;

	/* insure a pre-built packet is handled  */
	ps.s_txreq = get_waiting_verbs_txreq(ps.wait);
	do {
		/* Check for a constructed packet to be sent. */
		if (ps.s_txreq) {
			if (priv->s_flags & HFI1_S_TID_BUSY_SET) {
				qp->s_flags |= RVT_S_BUSY;
				ps.wait = iowait_get_ib_work(&priv->s_iowait);
			}
			spin_unlock_irqrestore(&qp->s_lock, ps.flags);

			/*
			 * If the packet cannot be sent now, return and
			 * the send tasklet will be woken up later.
			 */
			if (hfi1_verbs_send(qp, &ps))
				return;

			/* allow other tasks to run */
			if (hfi1_schedule_send_yield(qp, &ps, true))
				return;

			spin_lock_irqsave(&qp->s_lock, ps.flags);
			if (priv->s_flags & HFI1_S_TID_BUSY_SET) {
				qp->s_flags &= ~RVT_S_BUSY;
				priv->s_flags &= ~HFI1_S_TID_BUSY_SET;
				ps.wait = iowait_get_tid_work(&priv->s_iowait);
				if (iowait_flag_set(&priv->s_iowait,
						    IOWAIT_PENDING_IB))
					hfi1_schedule_send(qp);
			}
		}
	} while (hfi1_make_tid_rdma_pkt(qp, &ps));
	iowait_starve_clear(ps.pkts_sent, &priv->s_iowait);
	spin_unlock_irqrestore(&qp->s_lock, ps.flags);
}

static bool _hfi1_schedule_tid_send(struct rvt_qp *qp)
{
	struct hfi1_qp_priv *priv = qp->priv;
	struct hfi1_ibport *ibp =
		to_iport(qp->ibqp.device, qp->port_num);
	struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
	struct hfi1_devdata *dd = dd_from_ibdev(qp->ibqp.device);

	return iowait_tid_schedule(&priv->s_iowait, ppd->hfi1_wq,
				   priv->s_sde ?
				   priv->s_sde->cpu :
				   cpumask_first(cpumask_of_node(dd->node)));
}

/**
 * hfi1_schedule_tid_send - schedule progress on TID RDMA state machine
 * @qp: the QP
 *
 * This schedules qp progress on the TID RDMA state machine. Caller
 * should hold the s_lock.
 * Unlike hfi1_schedule_send(), this cannot use hfi1_send_ok() because
 * the two state machines can step on each other with respect to the
 * RVT_S_BUSY flag.
 * Therefore, a modified test is used.
 * @return true if the second leg is scheduled;
 *  false if the second leg is not scheduled.
 */
bool hfi1_schedule_tid_send(struct rvt_qp *qp)
{
	lockdep_assert_held(&qp->s_lock);
	if (hfi1_send_tid_ok(qp)) {
		/*
		 * The following call returns true if the qp is not on the
		 * queue and false if the qp is already on the queue before
		 * this call. Either way, the qp will be on the queue when the
		 * call returns.
		 */
		_hfi1_schedule_tid_send(qp);
		return true;
	}
	if (qp->s_flags & HFI1_S_ANY_WAIT_IO)
		iowait_set_flag(&((struct hfi1_qp_priv *)qp->priv)->s_iowait,
				IOWAIT_PENDING_TID);
	return false;
}

bool hfi1_tid_rdma_ack_interlock(struct rvt_qp *qp, struct rvt_ack_entry *e)
{
	struct rvt_ack_entry *prev;
	struct tid_rdma_request *req;
	struct hfi1_ibdev *dev = to_idev(qp->ibqp.device);
	struct hfi1_qp_priv *priv = qp->priv;
	u32 s_prev;

	s_prev = qp->s_tail_ack_queue == 0 ? rvt_size_atomic(&dev->rdi) :
		(qp->s_tail_ack_queue - 1);
	prev = &qp->s_ack_queue[s_prev];

	if ((e->opcode == TID_OP(READ_REQ) ||
	     e->opcode == OP(RDMA_READ_REQUEST)) &&
	    prev->opcode == TID_OP(WRITE_REQ)) {
		req = ack_to_tid_req(prev);
		if (req->ack_seg != req->total_segs) {
			priv->s_flags |= HFI1_R_TID_WAIT_INTERLCK;
			return true;
		}
	}
	return false;
}

static u32 read_r_next_psn(struct hfi1_devdata *dd, u8 ctxt, u8 fidx)
{
	u64 reg;

	/*
	 * The only sane way to get the amount of
	 * progress is to read the HW flow state.
	 */
	reg = read_uctxt_csr(dd, ctxt, RCV_TID_FLOW_TABLE + (8 * fidx));
	return mask_psn(reg);
}

static void tid_rdma_rcv_err(struct hfi1_packet *packet,
			     struct ib_other_headers *ohdr,
			     struct rvt_qp *qp, u32 psn, int diff, bool fecn)
{
	unsigned long flags;

	tid_rdma_rcv_error(packet, ohdr, qp, psn, diff);
	if (fecn) {
		spin_lock_irqsave(&qp->s_lock, flags);
		qp->s_flags |= RVT_S_ECN;
		spin_unlock_irqrestore(&qp->s_lock, flags);
	}
}

static void update_r_next_psn_fecn(struct hfi1_packet *packet,
				   struct hfi1_qp_priv *priv,
				   struct hfi1_ctxtdata *rcd,
				   struct tid_rdma_flow *flow,
				   bool fecn)
{
	/*
	 * If a start/middle packet is delivered here due to
	 * RSM rule and FECN, we need to update the r_next_psn.
	 */
	if (fecn && packet->etype == RHF_RCV_TYPE_EAGER &&
	    !(priv->s_flags & HFI1_R_TID_SW_PSN)) {
		struct hfi1_devdata *dd = rcd->dd;

		flow->flow_state.r_next_psn =
			read_r_next_psn(dd, rcd->ctxt, flow->idx);
	}
}