summaryrefslogtreecommitdiffstats
path: root/drivers/infiniband/hw/hfi1/rc.c
blob: 5da190e6011b1390f7861b2bd89a5a09e7976c14 (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
/*
 * Copyright(c) 2015, 2016 Intel Corporation.
 *
 * This file is provided under a dual BSD/GPLv2 license.  When using or
 * redistributing this file, you may do so under either license.
 *
 * GPL LICENSE SUMMARY
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of version 2 of the GNU General Public License as
 * published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful, but
 * WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 * BSD LICENSE
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 *
 *  - Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 *  - Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in
 *    the documentation and/or other materials provided with the
 *    distribution.
 *  - Neither the name of Intel Corporation nor the names of its
 *    contributors may be used to endorse or promote products derived
 *    from this software without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *
 */

#include <linux/io.h>
#include <rdma/rdma_vt.h>
#include <rdma/rdmavt_qp.h>

#include "hfi.h"
#include "qp.h"
#include "verbs_txreq.h"
#include "trace.h"

/* cut down ridiculously long IB macro names */
#define OP(x) IB_OPCODE_RC_##x

/**
 * hfi1_add_retry_timer - add/start a retry timer
 * @qp - the QP
 *
 * add a retry timer on the QP
 */
static inline void hfi1_add_retry_timer(struct rvt_qp *qp)
{
	struct ib_qp *ibqp = &qp->ibqp;
	struct rvt_dev_info *rdi = ib_to_rvt(ibqp->device);

	qp->s_flags |= RVT_S_TIMER;
	/* 4.096 usec. * (1 << qp->timeout) */
	qp->s_timer.expires = jiffies + qp->timeout_jiffies +
			      rdi->busy_jiffies;
	add_timer(&qp->s_timer);
}

/**
 * hfi1_add_rnr_timer - add/start an rnr timer
 * @qp - the QP
 * @to - timeout in usecs
 *
 * add an rnr timer on the QP
 */
void hfi1_add_rnr_timer(struct rvt_qp *qp, u32 to)
{
	struct hfi1_qp_priv *priv = qp->priv;

	qp->s_flags |= RVT_S_WAIT_RNR;
	qp->s_timer.expires = jiffies + usecs_to_jiffies(to);
	add_timer(&priv->s_rnr_timer);
}

/**
 * hfi1_mod_retry_timer - mod a retry timer
 * @qp - the QP
 *
 * Modify a potentially already running retry
 * timer
 */
static inline void hfi1_mod_retry_timer(struct rvt_qp *qp)
{
	struct ib_qp *ibqp = &qp->ibqp;
	struct rvt_dev_info *rdi = ib_to_rvt(ibqp->device);

	qp->s_flags |= RVT_S_TIMER;
	/* 4.096 usec. * (1 << qp->timeout) */
	mod_timer(&qp->s_timer, jiffies + qp->timeout_jiffies +
		  rdi->busy_jiffies);
}

/**
 * hfi1_stop_retry_timer - stop a retry timer
 * @qp - the QP
 *
 * stop a retry timer and return if the timer
 * had been pending.
 */
static inline int hfi1_stop_retry_timer(struct rvt_qp *qp)
{
	int rval = 0;

	/* Remove QP from retry */
	if (qp->s_flags & RVT_S_TIMER) {
		qp->s_flags &= ~RVT_S_TIMER;
		rval = del_timer(&qp->s_timer);
	}
	return rval;
}

/**
 * hfi1_stop_rc_timers - stop all timers
 * @qp - the QP
 *
 * stop any pending timers
 */
void hfi1_stop_rc_timers(struct rvt_qp *qp)
{
	struct hfi1_qp_priv *priv = qp->priv;

	/* Remove QP from all timers */
	if (qp->s_flags & (RVT_S_TIMER | RVT_S_WAIT_RNR)) {
		qp->s_flags &= ~(RVT_S_TIMER | RVT_S_WAIT_RNR);
		del_timer(&qp->s_timer);
		del_timer(&priv->s_rnr_timer);
	}
}

/**
 * hfi1_stop_rnr_timer - stop an rnr timer
 * @qp - the QP
 *
 * stop an rnr timer and return if the timer
 * had been pending.
 */
static inline int hfi1_stop_rnr_timer(struct rvt_qp *qp)
{
	int rval = 0;
	struct hfi1_qp_priv *priv = qp->priv;

	/* Remove QP from rnr timer */
	if (qp->s_flags & RVT_S_WAIT_RNR) {
		qp->s_flags &= ~RVT_S_WAIT_RNR;
		rval = del_timer(&priv->s_rnr_timer);
	}
	return rval;
}

/**
 * hfi1_del_timers_sync - wait for any timeout routines to exit
 * @qp - the QP
 */
void hfi1_del_timers_sync(struct rvt_qp *qp)
{
	struct hfi1_qp_priv *priv = qp->priv;

	del_timer_sync(&qp->s_timer);
	del_timer_sync(&priv->s_rnr_timer);
}

/* only opcode mask for adaptive pio */
const u32 rc_only_opcode =
	BIT(OP(SEND_ONLY) & 0x1f) |
	BIT(OP(SEND_ONLY_WITH_IMMEDIATE & 0x1f)) |
	BIT(OP(RDMA_WRITE_ONLY & 0x1f)) |
	BIT(OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE & 0x1f)) |
	BIT(OP(RDMA_READ_REQUEST & 0x1f)) |
	BIT(OP(ACKNOWLEDGE & 0x1f)) |
	BIT(OP(ATOMIC_ACKNOWLEDGE & 0x1f)) |
	BIT(OP(COMPARE_SWAP & 0x1f)) |
	BIT(OP(FETCH_ADD & 0x1f));

static u32 restart_sge(struct rvt_sge_state *ss, struct rvt_swqe *wqe,
		       u32 psn, u32 pmtu)
{
	u32 len;

	len = delta_psn(psn, wqe->psn) * pmtu;
	ss->sge = wqe->sg_list[0];
	ss->sg_list = wqe->sg_list + 1;
	ss->num_sge = wqe->wr.num_sge;
	ss->total_len = wqe->length;
	hfi1_skip_sge(ss, len, 0);
	return wqe->length - len;
}

/**
 * make_rc_ack - construct a response packet (ACK, NAK, or RDMA read)
 * @dev: the device for this QP
 * @qp: a pointer to the QP
 * @ohdr: a pointer to the IB header being constructed
 * @ps: the xmit packet state
 *
 * Return 1 if constructed; otherwise, return 0.
 * Note that we are in the responder's side of the QP context.
 * Note the QP s_lock must be held.
 */
static int make_rc_ack(struct hfi1_ibdev *dev, struct rvt_qp *qp,
		       struct hfi1_other_headers *ohdr,
		       struct hfi1_pkt_state *ps)
{
	struct rvt_ack_entry *e;
	u32 hwords;
	u32 len;
	u32 bth0;
	u32 bth2;
	int middle = 0;
	u32 pmtu = qp->pmtu;
	struct hfi1_qp_priv *priv = qp->priv;

	/* 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;

	switch (qp->s_ack_state) {
	case OP(RDMA_READ_RESPONSE_LAST):
	case OP(RDMA_READ_RESPONSE_ONLY):
		e = &qp->s_ack_queue[qp->s_tail_ack_queue];
		if (e->rdma_sge.mr) {
			rvt_put_mr(e->rdma_sge.mr);
			e->rdma_sge.mr = NULL;
		}
		/* FALLTHROUGH */
	case OP(ATOMIC_ACKNOWLEDGE):
		/*
		 * We can increment the tail pointer now that the last
		 * response has been sent instead of only being
		 * constructed.
		 */
		if (++qp->s_tail_ack_queue > HFI1_MAX_RDMA_ATOMIC)
			qp->s_tail_ack_queue = 0;
		/* FALLTHROUGH */
	case OP(SEND_ONLY):
	case OP(ACKNOWLEDGE):
		/* Check for no next entry in the queue. */
		if (qp->r_head_ack_queue == qp->s_tail_ack_queue) {
			if (qp->s_flags & RVT_S_ACK_PENDING)
				goto normal;
			goto bail;
		}

		e = &qp->s_ack_queue[qp->s_tail_ack_queue];
		if (e->opcode == OP(RDMA_READ_REQUEST)) {
			/*
			 * If a RDMA read response is being resent and
			 * we haven't seen the duplicate request yet,
			 * then stop sending the remaining responses the
			 * responder has seen until the requester re-sends it.
			 */
			len = e->rdma_sge.sge_length;
			if (len && !e->rdma_sge.mr) {
				qp->s_tail_ack_queue = qp->r_head_ack_queue;
				goto bail;
			}
			/* Copy SGE state in case we need to resend */
			ps->s_txreq->mr = e->rdma_sge.mr;
			if (ps->s_txreq->mr)
				rvt_get_mr(ps->s_txreq->mr);
			qp->s_ack_rdma_sge.sge = e->rdma_sge;
			qp->s_ack_rdma_sge.num_sge = 1;
			qp->s_cur_sge = &qp->s_ack_rdma_sge;
			if (len > pmtu) {
				len = pmtu;
				qp->s_ack_state = OP(RDMA_READ_RESPONSE_FIRST);
			} else {
				qp->s_ack_state = OP(RDMA_READ_RESPONSE_ONLY);
				e->sent = 1;
			}
			ohdr->u.aeth = hfi1_compute_aeth(qp);
			hwords++;
			qp->s_ack_rdma_psn = e->psn;
			bth2 = mask_psn(qp->s_ack_rdma_psn++);
		} else {
			/* COMPARE_SWAP or FETCH_ADD */
			qp->s_cur_sge = NULL;
			len = 0;
			qp->s_ack_state = OP(ATOMIC_ACKNOWLEDGE);
			ohdr->u.at.aeth = hfi1_compute_aeth(qp);
			ohdr->u.at.atomic_ack_eth[0] =
				cpu_to_be32(e->atomic_data >> 32);
			ohdr->u.at.atomic_ack_eth[1] =
				cpu_to_be32(e->atomic_data);
			hwords += sizeof(ohdr->u.at) / sizeof(u32);
			bth2 = mask_psn(e->psn);
			e->sent = 1;
		}
		bth0 = qp->s_ack_state << 24;
		break;

	case OP(RDMA_READ_RESPONSE_FIRST):
		qp->s_ack_state = OP(RDMA_READ_RESPONSE_MIDDLE);
		/* FALLTHROUGH */
	case OP(RDMA_READ_RESPONSE_MIDDLE):
		qp->s_cur_sge = &qp->s_ack_rdma_sge;
		ps->s_txreq->mr = qp->s_ack_rdma_sge.sge.mr;
		if (ps->s_txreq->mr)
			rvt_get_mr(ps->s_txreq->mr);
		len = qp->s_ack_rdma_sge.sge.sge_length;
		if (len > pmtu) {
			len = pmtu;
			middle = HFI1_CAP_IS_KSET(SDMA_AHG);
		} else {
			ohdr->u.aeth = hfi1_compute_aeth(qp);
			hwords++;
			qp->s_ack_state = OP(RDMA_READ_RESPONSE_LAST);
			e = &qp->s_ack_queue[qp->s_tail_ack_queue];
			e->sent = 1;
		}
		bth0 = qp->s_ack_state << 24;
		bth2 = mask_psn(qp->s_ack_rdma_psn++);
		break;

	default:
normal:
		/*
		 * Send a regular ACK.
		 * Set the s_ack_state so we wait until after sending
		 * the ACK before setting s_ack_state to ACKNOWLEDGE
		 * (see above).
		 */
		qp->s_ack_state = OP(SEND_ONLY);
		qp->s_flags &= ~RVT_S_ACK_PENDING;
		qp->s_cur_sge = NULL;
		if (qp->s_nak_state)
			ohdr->u.aeth =
				cpu_to_be32((qp->r_msn & HFI1_MSN_MASK) |
					    (qp->s_nak_state <<
					     HFI1_AETH_CREDIT_SHIFT));
		else
			ohdr->u.aeth = hfi1_compute_aeth(qp);
		hwords++;
		len = 0;
		bth0 = OP(ACKNOWLEDGE) << 24;
		bth2 = mask_psn(qp->s_ack_psn);
	}
	qp->s_rdma_ack_cnt++;
	qp->s_hdrwords = hwords;
	ps->s_txreq->sde = priv->s_sde;
	qp->s_cur_size = len;
	hfi1_make_ruc_header(qp, ohdr, bth0, bth2, middle, ps);
	/* pbc */
	ps->s_txreq->hdr_dwords = qp->s_hdrwords + 2;
	return 1;

bail:
	qp->s_ack_state = OP(ACKNOWLEDGE);
	/*
	 * Ensure s_rdma_ack_cnt changes are committed prior to resetting
	 * RVT_S_RESP_PENDING
	 */
	smp_wmb();
	qp->s_flags &= ~(RVT_S_RESP_PENDING
				| RVT_S_ACK_PENDING
				| RVT_S_AHG_VALID);
	return 0;
}

/**
 * hfi1_make_rc_req - construct a request packet (SEND, RDMA r/w, ATOMIC)
 * @qp: a pointer to the QP
 *
 * Assumes s_lock is held.
 *
 * Return 1 if constructed; otherwise, return 0.
 */
int hfi1_make_rc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps)
{
	struct hfi1_qp_priv *priv = qp->priv;
	struct hfi1_ibdev *dev = to_idev(qp->ibqp.device);
	struct hfi1_other_headers *ohdr;
	struct rvt_sge_state *ss;
	struct rvt_swqe *wqe;
	/* header size in 32-bit words LRH+BTH = (8+12)/4. */
	u32 hwords = 5;
	u32 len;
	u32 bth0 = 0;
	u32 bth2;
	u32 pmtu = qp->pmtu;
	char newreq;
	int middle = 0;
	int delta;

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

	ohdr = &ps->s_txreq->phdr.hdr.u.oth;
	if (qp->remote_ah_attr.ah_flags & IB_AH_GRH)
		ohdr = &ps->s_txreq->phdr.hdr.u.l.oth;

	/* Sending responses has higher priority over sending requests. */
	if ((qp->s_flags & RVT_S_RESP_PENDING) &&
	    make_rc_ack(dev, qp, ohdr, ps))
		return 1;

	if (!(ib_rvt_state_ops[qp->state] & RVT_PROCESS_SEND_OK)) {
		if (!(ib_rvt_state_ops[qp->state] & RVT_FLUSH_SEND))
			goto bail;
		/* We are in the error state, flush the work request. */
		smp_read_barrier_depends(); /* see post_one_send() */
		if (qp->s_last == ACCESS_ONCE(qp->s_head))
			goto bail;
		/* If DMAs are in progress, we can't flush immediately. */
		if (iowait_sdma_pending(&priv->s_iowait)) {
			qp->s_flags |= RVT_S_WAIT_DMA;
			goto bail;
		}
		clear_ahg(qp);
		wqe = rvt_get_swqe_ptr(qp, qp->s_last);
		hfi1_send_complete(qp, wqe, qp->s_last != qp->s_acked ?
			IB_WC_SUCCESS : IB_WC_WR_FLUSH_ERR);
		/* will get called again */
		goto done_free_tx;
	}

	if (qp->s_flags & (RVT_S_WAIT_RNR | RVT_S_WAIT_ACK))
		goto bail;

	if (cmp_psn(qp->s_psn, qp->s_sending_hpsn) <= 0) {
		if (cmp_psn(qp->s_sending_psn, qp->s_sending_hpsn) <= 0) {
			qp->s_flags |= RVT_S_WAIT_PSN;
			goto bail;
		}
		qp->s_sending_psn = qp->s_psn;
		qp->s_sending_hpsn = qp->s_psn - 1;
	}

	/* Send a request. */
	wqe = rvt_get_swqe_ptr(qp, qp->s_cur);
	switch (qp->s_state) {
	default:
		if (!(ib_rvt_state_ops[qp->state] & RVT_PROCESS_NEXT_SEND_OK))
			goto bail;
		/*
		 * Resend an old request or start a new one.
		 *
		 * We keep track of the current SWQE so that
		 * we don't reset the "furthest progress" state
		 * if we need to back up.
		 */
		newreq = 0;
		if (qp->s_cur == qp->s_tail) {
			/* Check if send work queue is empty. */
			if (qp->s_tail == qp->s_head) {
				clear_ahg(qp);
				goto bail;
			}
			/*
			 * If a fence is requested, wait for previous
			 * RDMA read and atomic operations to finish.
			 */
			if ((wqe->wr.send_flags & IB_SEND_FENCE) &&
			    qp->s_num_rd_atomic) {
				qp->s_flags |= RVT_S_WAIT_FENCE;
				goto bail;
			}
			/*
			 * Local operations are processed immediately
			 * after all prior requests have completed
			 */
			if (wqe->wr.opcode == IB_WR_REG_MR ||
			    wqe->wr.opcode == IB_WR_LOCAL_INV) {
				int local_ops = 0;
				int err = 0;

				if (qp->s_last != qp->s_cur)
					goto bail;
				if (++qp->s_cur == qp->s_size)
					qp->s_cur = 0;
				if (++qp->s_tail == qp->s_size)
					qp->s_tail = 0;
				if (!(wqe->wr.send_flags &
				      RVT_SEND_COMPLETION_ONLY)) {
					err = rvt_invalidate_rkey(
						qp,
						wqe->wr.ex.invalidate_rkey);
					local_ops = 1;
				}
				hfi1_send_complete(qp, wqe,
						   err ? IB_WC_LOC_PROT_ERR
						       : IB_WC_SUCCESS);
				if (local_ops)
					atomic_dec(&qp->local_ops_pending);
				qp->s_hdrwords = 0;
				goto done_free_tx;
			}

			newreq = 1;
			qp->s_psn = wqe->psn;
		}
		/*
		 * Note that we have to be careful not to modify the
		 * original work request since we may need to resend
		 * it.
		 */
		len = wqe->length;
		ss = &qp->s_sge;
		bth2 = mask_psn(qp->s_psn);
		switch (wqe->wr.opcode) {
		case IB_WR_SEND:
		case IB_WR_SEND_WITH_IMM:
		case IB_WR_SEND_WITH_INV:
			/* If no credit, return. */
			if (!(qp->s_flags & RVT_S_UNLIMITED_CREDIT) &&
			    cmp_msn(wqe->ssn, qp->s_lsn + 1) > 0) {
				qp->s_flags |= RVT_S_WAIT_SSN_CREDIT;
				goto bail;
			}
			if (len > pmtu) {
				qp->s_state = OP(SEND_FIRST);
				len = pmtu;
				break;
			}
			if (wqe->wr.opcode == IB_WR_SEND) {
				qp->s_state = OP(SEND_ONLY);
			} else if (wqe->wr.opcode == IB_WR_SEND_WITH_IMM) {
				qp->s_state = OP(SEND_ONLY_WITH_IMMEDIATE);
				/* Immediate data comes after the BTH */
				ohdr->u.imm_data = wqe->wr.ex.imm_data;
				hwords += 1;
			} else {
				qp->s_state = OP(SEND_ONLY_WITH_INVALIDATE);
				/* Invalidate rkey comes after the BTH */
				ohdr->u.ieth = cpu_to_be32(
						wqe->wr.ex.invalidate_rkey);
				hwords += 1;
			}
			if (wqe->wr.send_flags & IB_SEND_SOLICITED)
				bth0 |= IB_BTH_SOLICITED;
			bth2 |= IB_BTH_REQ_ACK;
			if (++qp->s_cur == qp->s_size)
				qp->s_cur = 0;
			break;

		case IB_WR_RDMA_WRITE:
			if (newreq && !(qp->s_flags & RVT_S_UNLIMITED_CREDIT))
				qp->s_lsn++;
			/* FALLTHROUGH */
		case IB_WR_RDMA_WRITE_WITH_IMM:
			/* If no credit, return. */
			if (!(qp->s_flags & RVT_S_UNLIMITED_CREDIT) &&
			    cmp_msn(wqe->ssn, qp->s_lsn + 1) > 0) {
				qp->s_flags |= RVT_S_WAIT_SSN_CREDIT;
				goto bail;
			}
			ohdr->u.rc.reth.vaddr =
				cpu_to_be64(wqe->rdma_wr.remote_addr);
			ohdr->u.rc.reth.rkey =
				cpu_to_be32(wqe->rdma_wr.rkey);
			ohdr->u.rc.reth.length = cpu_to_be32(len);
			hwords += sizeof(struct ib_reth) / sizeof(u32);
			if (len > pmtu) {
				qp->s_state = OP(RDMA_WRITE_FIRST);
				len = pmtu;
				break;
			}
			if (wqe->wr.opcode == IB_WR_RDMA_WRITE) {
				qp->s_state = OP(RDMA_WRITE_ONLY);
			} else {
				qp->s_state =
					OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE);
				/* Immediate data comes after RETH */
				ohdr->u.rc.imm_data = wqe->wr.ex.imm_data;
				hwords += 1;
				if (wqe->wr.send_flags & IB_SEND_SOLICITED)
					bth0 |= IB_BTH_SOLICITED;
			}
			bth2 |= IB_BTH_REQ_ACK;
			if (++qp->s_cur == qp->s_size)
				qp->s_cur = 0;
			break;

		case IB_WR_RDMA_READ:
			/*
			 * Don't allow more operations to be started
			 * than the QP limits allow.
			 */
			if (newreq) {
				if (qp->s_num_rd_atomic >=
				    qp->s_max_rd_atomic) {
					qp->s_flags |= RVT_S_WAIT_RDMAR;
					goto bail;
				}
				qp->s_num_rd_atomic++;
				if (!(qp->s_flags & RVT_S_UNLIMITED_CREDIT))
					qp->s_lsn++;
			}
			ohdr->u.rc.reth.vaddr =
				cpu_to_be64(wqe->rdma_wr.remote_addr);
			ohdr->u.rc.reth.rkey =
				cpu_to_be32(wqe->rdma_wr.rkey);
			ohdr->u.rc.reth.length = cpu_to_be32(len);
			qp->s_state = OP(RDMA_READ_REQUEST);
			hwords += sizeof(ohdr->u.rc.reth) / sizeof(u32);
			ss = NULL;
			len = 0;
			bth2 |= IB_BTH_REQ_ACK;
			if (++qp->s_cur == qp->s_size)
				qp->s_cur = 0;
			break;

		case IB_WR_ATOMIC_CMP_AND_SWP:
		case IB_WR_ATOMIC_FETCH_AND_ADD:
			/*
			 * Don't allow more operations to be started
			 * than the QP limits allow.
			 */
			if (newreq) {
				if (qp->s_num_rd_atomic >=
				    qp->s_max_rd_atomic) {
					qp->s_flags |= RVT_S_WAIT_RDMAR;
					goto bail;
				}
				qp->s_num_rd_atomic++;
				if (!(qp->s_flags & RVT_S_UNLIMITED_CREDIT))
					qp->s_lsn++;
			}
			if (wqe->wr.opcode == IB_WR_ATOMIC_CMP_AND_SWP) {
				qp->s_state = OP(COMPARE_SWAP);
				ohdr->u.atomic_eth.swap_data = cpu_to_be64(
					wqe->atomic_wr.swap);
				ohdr->u.atomic_eth.compare_data = cpu_to_be64(
					wqe->atomic_wr.compare_add);
			} else {
				qp->s_state = OP(FETCH_ADD);
				ohdr->u.atomic_eth.swap_data = cpu_to_be64(
					wqe->atomic_wr.compare_add);
				ohdr->u.atomic_eth.compare_data = 0;
			}
			ohdr->u.atomic_eth.vaddr[0] = cpu_to_be32(
				wqe->atomic_wr.remote_addr >> 32);
			ohdr->u.atomic_eth.vaddr[1] = cpu_to_be32(
				wqe->atomic_wr.remote_addr);
			ohdr->u.atomic_eth.rkey = cpu_to_be32(
				wqe->atomic_wr.rkey);
			hwords += sizeof(struct ib_atomic_eth) / sizeof(u32);
			ss = NULL;
			len = 0;
			bth2 |= IB_BTH_REQ_ACK;
			if (++qp->s_cur == qp->s_size)
				qp->s_cur = 0;
			break;

		default:
			goto bail;
		}
		qp->s_sge.sge = wqe->sg_list[0];
		qp->s_sge.sg_list = wqe->sg_list + 1;
		qp->s_sge.num_sge = wqe->wr.num_sge;
		qp->s_sge.total_len = wqe->length;
		qp->s_len = wqe->length;
		if (newreq) {
			qp->s_tail++;
			if (qp->s_tail >= qp->s_size)
				qp->s_tail = 0;
		}
		if (wqe->wr.opcode == IB_WR_RDMA_READ)
			qp->s_psn = wqe->lpsn + 1;
		else
			qp->s_psn++;
		break;

	case OP(RDMA_READ_RESPONSE_FIRST):
		/*
		 * qp->s_state is normally set to the opcode of the
		 * last packet constructed for new requests and therefore
		 * is never set to RDMA read response.
		 * RDMA_READ_RESPONSE_FIRST is used by the ACK processing
		 * thread to indicate a SEND needs to be restarted from an
		 * earlier PSN without interfering with the sending thread.
		 * See restart_rc().
		 */
		qp->s_len = restart_sge(&qp->s_sge, wqe, qp->s_psn, pmtu);
		/* FALLTHROUGH */
	case OP(SEND_FIRST):
		qp->s_state = OP(SEND_MIDDLE);
		/* FALLTHROUGH */
	case OP(SEND_MIDDLE):
		bth2 = mask_psn(qp->s_psn++);
		ss = &qp->s_sge;
		len = qp->s_len;
		if (len > pmtu) {
			len = pmtu;
			middle = HFI1_CAP_IS_KSET(SDMA_AHG);
			break;
		}
		if (wqe->wr.opcode == IB_WR_SEND) {
			qp->s_state = OP(SEND_LAST);
		} else if (wqe->wr.opcode == IB_WR_SEND_WITH_IMM) {
			qp->s_state = OP(SEND_LAST_WITH_IMMEDIATE);
			/* Immediate data comes after the BTH */
			ohdr->u.imm_data = wqe->wr.ex.imm_data;
			hwords += 1;
		} else {
			qp->s_state = OP(SEND_LAST_WITH_INVALIDATE);
			/* invalidate data comes after the BTH */
			ohdr->u.ieth = cpu_to_be32(wqe->wr.ex.invalidate_rkey);
			hwords += 1;
		}
		if (wqe->wr.send_flags & IB_SEND_SOLICITED)
			bth0 |= IB_BTH_SOLICITED;
		bth2 |= IB_BTH_REQ_ACK;
		qp->s_cur++;
		if (qp->s_cur >= qp->s_size)
			qp->s_cur = 0;
		break;

	case OP(RDMA_READ_RESPONSE_LAST):
		/*
		 * qp->s_state is normally set to the opcode of the
		 * last packet constructed for new requests and therefore
		 * is never set to RDMA read response.
		 * RDMA_READ_RESPONSE_LAST is used by the ACK processing
		 * thread to indicate a RDMA write needs to be restarted from
		 * an earlier PSN without interfering with the sending thread.
		 * See restart_rc().
		 */
		qp->s_len = restart_sge(&qp->s_sge, wqe, qp->s_psn, pmtu);
		/* FALLTHROUGH */
	case OP(RDMA_WRITE_FIRST):
		qp->s_state = OP(RDMA_WRITE_MIDDLE);
		/* FALLTHROUGH */
	case OP(RDMA_WRITE_MIDDLE):
		bth2 = mask_psn(qp->s_psn++);
		ss = &qp->s_sge;
		len = qp->s_len;
		if (len > pmtu) {
			len = pmtu;
			middle = HFI1_CAP_IS_KSET(SDMA_AHG);
			break;
		}
		if (wqe->wr.opcode == IB_WR_RDMA_WRITE) {
			qp->s_state = OP(RDMA_WRITE_LAST);
		} else {
			qp->s_state = OP(RDMA_WRITE_LAST_WITH_IMMEDIATE);
			/* Immediate data comes after the BTH */
			ohdr->u.imm_data = wqe->wr.ex.imm_data;
			hwords += 1;
			if (wqe->wr.send_flags & IB_SEND_SOLICITED)
				bth0 |= IB_BTH_SOLICITED;
		}
		bth2 |= IB_BTH_REQ_ACK;
		qp->s_cur++;
		if (qp->s_cur >= qp->s_size)
			qp->s_cur = 0;
		break;

	case OP(RDMA_READ_RESPONSE_MIDDLE):
		/*
		 * qp->s_state is normally set to the opcode of the
		 * last packet constructed for new requests and therefore
		 * is never set to RDMA read response.
		 * RDMA_READ_RESPONSE_MIDDLE is used by the ACK processing
		 * thread to indicate a RDMA read needs to be restarted from
		 * an earlier PSN without interfering with the sending thread.
		 * See restart_rc().
		 */
		len = (delta_psn(qp->s_psn, wqe->psn)) * pmtu;
		ohdr->u.rc.reth.vaddr =
			cpu_to_be64(wqe->rdma_wr.remote_addr + len);
		ohdr->u.rc.reth.rkey =
			cpu_to_be32(wqe->rdma_wr.rkey);
		ohdr->u.rc.reth.length = cpu_to_be32(wqe->length - len);
		qp->s_state = OP(RDMA_READ_REQUEST);
		hwords += sizeof(ohdr->u.rc.reth) / sizeof(u32);
		bth2 = mask_psn(qp->s_psn) | IB_BTH_REQ_ACK;
		qp->s_psn = wqe->lpsn + 1;
		ss = NULL;
		len = 0;
		qp->s_cur++;
		if (qp->s_cur == qp->s_size)
			qp->s_cur = 0;
		break;
	}
	qp->s_sending_hpsn = bth2;
	delta = delta_psn(bth2, wqe->psn);
	if (delta && delta % HFI1_PSN_CREDIT == 0)
		bth2 |= IB_BTH_REQ_ACK;
	if (qp->s_flags & RVT_S_SEND_ONE) {
		qp->s_flags &= ~RVT_S_SEND_ONE;
		qp->s_flags |= RVT_S_WAIT_ACK;
		bth2 |= IB_BTH_REQ_ACK;
	}
	qp->s_len -= len;
	qp->s_hdrwords = hwords;
	ps->s_txreq->sde = priv->s_sde;
	qp->s_cur_sge = ss;
	qp->s_cur_size = len;
	hfi1_make_ruc_header(
		qp,
		ohdr,
		bth0 | (qp->s_state << 24),
		bth2,
		middle,
		ps);
	/* pbc */
	ps->s_txreq->hdr_dwords = qp->s_hdrwords + 2;
	return 1;

done_free_tx:
	hfi1_put_txreq(ps->s_txreq);
	ps->s_txreq = NULL;
	return 1;

bail:
	hfi1_put_txreq(ps->s_txreq);

bail_no_tx:
	ps->s_txreq = NULL;
	qp->s_flags &= ~RVT_S_BUSY;
	qp->s_hdrwords = 0;
	return 0;
}

/**
 * hfi1_send_rc_ack - Construct an ACK packet and send it
 * @qp: a pointer to the QP
 *
 * This is called from hfi1_rc_rcv() and handle_receive_interrupt().
 * Note that RDMA reads and atomics are handled in the
 * send side QP state and tasklet.
 */
void hfi1_send_rc_ack(struct hfi1_ctxtdata *rcd, struct rvt_qp *qp,
		      int is_fecn)
{
	struct hfi1_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
	struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
	u64 pbc, pbc_flags = 0;
	u16 lrh0;
	u16 sc5;
	u32 bth0;
	u32 hwords;
	u32 vl, plen;
	struct send_context *sc;
	struct pio_buf *pbuf;
	struct hfi1_ib_header hdr;
	struct hfi1_other_headers *ohdr;
	unsigned long flags;

	/* Don't send ACK or NAK if a RDMA read or atomic is pending. */
	if (qp->s_flags & RVT_S_RESP_PENDING)
		goto queue_ack;

	/* Ensure s_rdma_ack_cnt changes are committed */
	smp_read_barrier_depends();
	if (qp->s_rdma_ack_cnt)
		goto queue_ack;

	/* Construct the header */
	/* header size in 32-bit words LRH+BTH+AETH = (8+12+4)/4 */
	hwords = 6;
	if (unlikely(qp->remote_ah_attr.ah_flags & IB_AH_GRH)) {
		hwords += hfi1_make_grh(ibp, &hdr.u.l.grh,
				       &qp->remote_ah_attr.grh, hwords, 0);
		ohdr = &hdr.u.l.oth;
		lrh0 = HFI1_LRH_GRH;
	} else {
		ohdr = &hdr.u.oth;
		lrh0 = HFI1_LRH_BTH;
	}
	/* read pkey_index w/o lock (its atomic) */
	bth0 = hfi1_get_pkey(ibp, qp->s_pkey_index) | (OP(ACKNOWLEDGE) << 24);
	if (qp->s_mig_state == IB_MIG_MIGRATED)
		bth0 |= IB_BTH_MIG_REQ;
	if (qp->r_nak_state)
		ohdr->u.aeth = cpu_to_be32((qp->r_msn & HFI1_MSN_MASK) |
					    (qp->r_nak_state <<
					     HFI1_AETH_CREDIT_SHIFT));
	else
		ohdr->u.aeth = hfi1_compute_aeth(qp);
	sc5 = ibp->sl_to_sc[qp->remote_ah_attr.sl];
	/* set PBC_DC_INFO bit (aka SC[4]) in pbc_flags */
	pbc_flags |= ((!!(sc5 & 0x10)) << PBC_DC_INFO_SHIFT);
	lrh0 |= (sc5 & 0xf) << 12 | (qp->remote_ah_attr.sl & 0xf) << 4;
	hdr.lrh[0] = cpu_to_be16(lrh0);
	hdr.lrh[1] = cpu_to_be16(qp->remote_ah_attr.dlid);
	hdr.lrh[2] = cpu_to_be16(hwords + SIZE_OF_CRC);
	hdr.lrh[3] = cpu_to_be16(ppd->lid | qp->remote_ah_attr.src_path_bits);
	ohdr->bth[0] = cpu_to_be32(bth0);
	ohdr->bth[1] = cpu_to_be32(qp->remote_qpn);
	ohdr->bth[1] |= cpu_to_be32((!!is_fecn) << HFI1_BECN_SHIFT);
	ohdr->bth[2] = cpu_to_be32(mask_psn(qp->r_ack_psn));

	/* Don't try to send ACKs if the link isn't ACTIVE */
	if (driver_lstate(ppd) != IB_PORT_ACTIVE)
		return;

	sc = rcd->sc;
	plen = 2 /* PBC */ + hwords;
	vl = sc_to_vlt(ppd->dd, sc5);
	pbc = create_pbc(ppd, pbc_flags, qp->srate_mbps, vl, plen);

	pbuf = sc_buffer_alloc(sc, plen, NULL, NULL);
	if (!pbuf) {
		/*
		 * We have no room to send at the moment.  Pass
		 * responsibility for sending the ACK to the send tasklet
		 * so that when enough buffer space becomes available,
		 * the ACK is sent ahead of other outgoing packets.
		 */
		goto queue_ack;
	}

	trace_ack_output_ibhdr(dd_from_ibdev(qp->ibqp.device), &hdr);

	/* write the pbc and data */
	ppd->dd->pio_inline_send(ppd->dd, pbuf, pbc, &hdr, hwords);

	return;

queue_ack:
	this_cpu_inc(*ibp->rvp.rc_qacks);
	spin_lock_irqsave(&qp->s_lock, flags);
	qp->s_flags |= RVT_S_ACK_PENDING | RVT_S_RESP_PENDING;
	qp->s_nak_state = qp->r_nak_state;
	qp->s_ack_psn = qp->r_ack_psn;
	if (is_fecn)
		qp->s_flags |= RVT_S_ECN;

	/* Schedule the send tasklet. */
	hfi1_schedule_send(qp);
	spin_unlock_irqrestore(&qp->s_lock, flags);
}

/**
 * reset_psn - reset the QP state to send starting from PSN
 * @qp: the QP
 * @psn: the packet sequence number to restart at
 *
 * This is called from hfi1_rc_rcv() to process an incoming RC ACK
 * for the given QP.
 * Called at interrupt level with the QP s_lock held.
 */
static void reset_psn(struct rvt_qp *qp, u32 psn)
{
	u32 n = qp->s_acked;
	struct rvt_swqe *wqe = rvt_get_swqe_ptr(qp, n);
	u32 opcode;

	qp->s_cur = n;

	/*
	 * If we are starting the request from the beginning,
	 * let the normal send code handle initialization.
	 */
	if (cmp_psn(psn, wqe->psn) <= 0) {
		qp->s_state = OP(SEND_LAST);
		goto done;
	}

	/* Find the work request opcode corresponding to the given PSN. */
	opcode = wqe->wr.opcode;
	for (;;) {
		int diff;

		if (++n == qp->s_size)
			n = 0;
		if (n == qp->s_tail)
			break;
		wqe = rvt_get_swqe_ptr(qp, n);
		diff = cmp_psn(psn, wqe->psn);
		if (diff < 0)
			break;
		qp->s_cur = n;
		/*
		 * If we are starting the request from the beginning,
		 * let the normal send code handle initialization.
		 */
		if (diff == 0) {
			qp->s_state = OP(SEND_LAST);
			goto done;
		}
		opcode = wqe->wr.opcode;
	}

	/*
	 * Set the state to restart in the middle of a request.
	 * Don't change the s_sge, s_cur_sge, or s_cur_size.
	 * See hfi1_make_rc_req().
	 */
	switch (opcode) {
	case IB_WR_SEND:
	case IB_WR_SEND_WITH_IMM:
		qp->s_state = OP(RDMA_READ_RESPONSE_FIRST);
		break;

	case IB_WR_RDMA_WRITE:
	case IB_WR_RDMA_WRITE_WITH_IMM:
		qp->s_state = OP(RDMA_READ_RESPONSE_LAST);
		break;

	case IB_WR_RDMA_READ:
		qp->s_state = OP(RDMA_READ_RESPONSE_MIDDLE);
		break;

	default:
		/*
		 * This case shouldn't happen since its only
		 * one PSN per req.
		 */
		qp->s_state = OP(SEND_LAST);
	}
done:
	qp->s_psn = psn;
	/*
	 * Set RVT_S_WAIT_PSN as rc_complete() may start the timer
	 * asynchronously before the send tasklet can get scheduled.
	 * Doing it in hfi1_make_rc_req() is too late.
	 */
	if ((cmp_psn(qp->s_psn, qp->s_sending_hpsn) <= 0) &&
	    (cmp_psn(qp->s_sending_psn, qp->s_sending_hpsn) <= 0))
		qp->s_flags |= RVT_S_WAIT_PSN;
	qp->s_flags &= ~RVT_S_AHG_VALID;
}

/*
 * Back up requester to resend the last un-ACKed request.
 * The QP r_lock and s_lock should be held and interrupts disabled.
 */
static void restart_rc(struct rvt_qp *qp, u32 psn, int wait)
{
	struct rvt_swqe *wqe = rvt_get_swqe_ptr(qp, qp->s_acked);
	struct hfi1_ibport *ibp;

	if (qp->s_retry == 0) {
		if (qp->s_mig_state == IB_MIG_ARMED) {
			hfi1_migrate_qp(qp);
			qp->s_retry = qp->s_retry_cnt;
		} else if (qp->s_last == qp->s_acked) {
			hfi1_send_complete(qp, wqe, IB_WC_RETRY_EXC_ERR);
			rvt_error_qp(qp, IB_WC_WR_FLUSH_ERR);
			return;
		} else { /* need to handle delayed completion */
			return;
		}
	} else {
		qp->s_retry--;
	}

	ibp = to_iport(qp->ibqp.device, qp->port_num);
	if (wqe->wr.opcode == IB_WR_RDMA_READ)
		ibp->rvp.n_rc_resends++;
	else
		ibp->rvp.n_rc_resends += delta_psn(qp->s_psn, psn);

	qp->s_flags &= ~(RVT_S_WAIT_FENCE | RVT_S_WAIT_RDMAR |
			 RVT_S_WAIT_SSN_CREDIT | RVT_S_WAIT_PSN |
			 RVT_S_WAIT_ACK);
	if (wait)
		qp->s_flags |= RVT_S_SEND_ONE;
	reset_psn(qp, psn);
}

/*
 * This is called from s_timer for missing responses.
 */
void hfi1_rc_timeout(unsigned long arg)
{
	struct rvt_qp *qp = (struct rvt_qp *)arg;
	struct hfi1_ibport *ibp;
	unsigned long flags;

	spin_lock_irqsave(&qp->r_lock, flags);
	spin_lock(&qp->s_lock);
	if (qp->s_flags & RVT_S_TIMER) {
		ibp = to_iport(qp->ibqp.device, qp->port_num);
		ibp->rvp.n_rc_timeouts++;
		qp->s_flags &= ~RVT_S_TIMER;
		del_timer(&qp->s_timer);
		trace_hfi1_timeout(qp, qp->s_last_psn + 1);
		restart_rc(qp, qp->s_last_psn + 1, 1);
		hfi1_schedule_send(qp);
	}
	spin_unlock(&qp->s_lock);
	spin_unlock_irqrestore(&qp->r_lock, flags);
}

/*
 * This is called from s_timer for RNR timeouts.
 */
void hfi1_rc_rnr_retry(unsigned long arg)
{
	struct rvt_qp *qp = (struct rvt_qp *)arg;
	unsigned long flags;

	spin_lock_irqsave(&qp->s_lock, flags);
	hfi1_stop_rnr_timer(qp);
	hfi1_schedule_send(qp);
	spin_unlock_irqrestore(&qp->s_lock, flags);
}

/*
 * Set qp->s_sending_psn to the next PSN after the given one.
 * This would be psn+1 except when RDMA reads are present.
 */
static void reset_sending_psn(struct rvt_qp *qp, u32 psn)
{
	struct rvt_swqe *wqe;
	u32 n = qp->s_last;

	/* Find the work request corresponding to the given PSN. */
	for (;;) {
		wqe = rvt_get_swqe_ptr(qp, n);
		if (cmp_psn(psn, wqe->lpsn) <= 0) {
			if (wqe->wr.opcode == IB_WR_RDMA_READ)
				qp->s_sending_psn = wqe->lpsn + 1;
			else
				qp->s_sending_psn = psn + 1;
			break;
		}
		if (++n == qp->s_size)
			n = 0;
		if (n == qp->s_tail)
			break;
	}
}

/*
 * This should be called with the QP s_lock held and interrupts disabled.
 */
void hfi1_rc_send_complete(struct rvt_qp *qp, struct hfi1_ib_header *hdr)
{
	struct hfi1_other_headers *ohdr;
	struct rvt_swqe *wqe;
	struct ib_wc wc;
	unsigned i;
	u32 opcode;
	u32 psn;

	if (!(ib_rvt_state_ops[qp->state] & RVT_PROCESS_OR_FLUSH_SEND))
		return;

	/* Find out where the BTH is */
	if ((be16_to_cpu(hdr->lrh[0]) & 3) == HFI1_LRH_BTH)
		ohdr = &hdr->u.oth;
	else
		ohdr = &hdr->u.l.oth;

	opcode = be32_to_cpu(ohdr->bth[0]) >> 24;
	if (opcode >= OP(RDMA_READ_RESPONSE_FIRST) &&
	    opcode <= OP(ATOMIC_ACKNOWLEDGE)) {
		WARN_ON(!qp->s_rdma_ack_cnt);
		qp->s_rdma_ack_cnt--;
		return;
	}

	psn = be32_to_cpu(ohdr->bth[2]);
	reset_sending_psn(qp, psn);

	/*
	 * Start timer after a packet requesting an ACK has been sent and
	 * there are still requests that haven't been acked.
	 */
	if ((psn & IB_BTH_REQ_ACK) && qp->s_acked != qp->s_tail &&
	    !(qp->s_flags &
		(RVT_S_TIMER | RVT_S_WAIT_RNR | RVT_S_WAIT_PSN)) &&
		(ib_rvt_state_ops[qp->state] & RVT_PROCESS_RECV_OK))
		hfi1_add_retry_timer(qp);

	while (qp->s_last != qp->s_acked) {
		u32 s_last;

		wqe = rvt_get_swqe_ptr(qp, qp->s_last);
		if (cmp_psn(wqe->lpsn, qp->s_sending_psn) >= 0 &&
		    cmp_psn(qp->s_sending_psn, qp->s_sending_hpsn) <= 0)
			break;
		s_last = qp->s_last;
		if (++s_last >= qp->s_size)
			s_last = 0;
		qp->s_last = s_last;
		/* see post_send() */
		barrier();
		for (i = 0; i < wqe->wr.num_sge; i++) {
			struct rvt_sge *sge = &wqe->sg_list[i];

			rvt_put_mr(sge->mr);
		}
		/* Post a send completion queue entry if requested. */
		if (!(qp->s_flags & RVT_S_SIGNAL_REQ_WR) ||
		    (wqe->wr.send_flags & IB_SEND_SIGNALED)) {
			memset(&wc, 0, sizeof(wc));
			wc.wr_id = wqe->wr.wr_id;
			wc.status = IB_WC_SUCCESS;
			wc.opcode = ib_hfi1_wc_opcode[wqe->wr.opcode];
			wc.byte_len = wqe->length;
			wc.qp = &qp->ibqp;
			rvt_cq_enter(ibcq_to_rvtcq(qp->ibqp.send_cq), &wc, 0);
		}
	}
	/*
	 * If we were waiting for sends to complete before re-sending,
	 * and they are now complete, restart sending.
	 */
	trace_hfi1_sendcomplete(qp, psn);
	if (qp->s_flags & RVT_S_WAIT_PSN &&
	    cmp_psn(qp->s_sending_psn, qp->s_sending_hpsn) > 0) {
		qp->s_flags &= ~RVT_S_WAIT_PSN;
		qp->s_sending_psn = qp->s_psn;
		qp->s_sending_hpsn = qp->s_psn - 1;
		hfi1_schedule_send(qp);
	}
}

static inline void update_last_psn(struct rvt_qp *qp, u32 psn)
{
	qp->s_last_psn = psn;
}

/*
 * Generate a SWQE completion.
 * This is similar to hfi1_send_complete but has to check to be sure
 * that the SGEs are not being referenced if the SWQE is being resent.
 */
static struct rvt_swqe *do_rc_completion(struct rvt_qp *qp,
					 struct rvt_swqe *wqe,
					 struct hfi1_ibport *ibp)
{
	struct ib_wc wc;
	unsigned i;

	/*
	 * Don't decrement refcount and don't generate a
	 * completion if the SWQE is being resent until the send
	 * is finished.
	 */
	if (cmp_psn(wqe->lpsn, qp->s_sending_psn) < 0 ||
	    cmp_psn(qp->s_sending_psn, qp->s_sending_hpsn) > 0) {
		u32 s_last;

		for (i = 0; i < wqe->wr.num_sge; i++) {
			struct rvt_sge *sge = &wqe->sg_list[i];

			rvt_put_mr(sge->mr);
		}
		s_last = qp->s_last;
		if (++s_last >= qp->s_size)
			s_last = 0;
		qp->s_last = s_last;
		/* see post_send() */
		barrier();
		/* Post a send completion queue entry if requested. */
		if (!(qp->s_flags & RVT_S_SIGNAL_REQ_WR) ||
		    (wqe->wr.send_flags & IB_SEND_SIGNALED)) {
			memset(&wc, 0, sizeof(wc));
			wc.wr_id = wqe->wr.wr_id;
			wc.status = IB_WC_SUCCESS;
			wc.opcode = ib_hfi1_wc_opcode[wqe->wr.opcode];
			wc.byte_len = wqe->length;
			wc.qp = &qp->ibqp;
			rvt_cq_enter(ibcq_to_rvtcq(qp->ibqp.send_cq), &wc, 0);
		}
	} else {
		struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);

		this_cpu_inc(*ibp->rvp.rc_delayed_comp);
		/*
		 * If send progress not running attempt to progress
		 * SDMA queue.
		 */
		if (ppd->dd->flags & HFI1_HAS_SEND_DMA) {
			struct sdma_engine *engine;
			u8 sc5;

			/* For now use sc to find engine */
			sc5 = ibp->sl_to_sc[qp->remote_ah_attr.sl];
			engine = qp_to_sdma_engine(qp, sc5);
			sdma_engine_progress_schedule(engine);
		}
	}

	qp->s_retry = qp->s_retry_cnt;
	update_last_psn(qp, wqe->lpsn);

	/*
	 * If we are completing a request which is in the process of
	 * being resent, we can stop re-sending it since we know the
	 * responder has already seen it.
	 */
	if (qp->s_acked == qp->s_cur) {
		if (++qp->s_cur >= qp->s_size)
			qp->s_cur = 0;
		qp->s_acked = qp->s_cur;
		wqe = rvt_get_swqe_ptr(qp, qp->s_cur);
		if (qp->s_acked != qp->s_tail) {
			qp->s_state = OP(SEND_LAST);
			qp->s_psn = wqe->psn;
		}
	} else {
		if (++qp->s_acked >= qp->s_size)
			qp->s_acked = 0;
		if (qp->state == IB_QPS_SQD && qp->s_acked == qp->s_cur)
			qp->s_draining = 0;
		wqe = rvt_get_swqe_ptr(qp, qp->s_acked);
	}
	return wqe;
}

/**
 * do_rc_ack - process an incoming RC ACK
 * @qp: the QP the ACK came in on
 * @psn: the packet sequence number of the ACK
 * @opcode: the opcode of the request that resulted in the ACK
 *
 * This is called from rc_rcv_resp() to process an incoming RC ACK
 * for the given QP.
 * May be called at interrupt level, with the QP s_lock held.
 * Returns 1 if OK, 0 if current operation should be aborted (NAK).
 */
static int do_rc_ack(struct rvt_qp *qp, u32 aeth, u32 psn, int opcode,
		     u64 val, struct hfi1_ctxtdata *rcd)
{
	struct hfi1_ibport *ibp;
	enum ib_wc_status status;
	struct rvt_swqe *wqe;
	int ret = 0;
	u32 ack_psn;
	int diff;
	unsigned long to;

	/*
	 * 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.  The MSN won't include the NAK'ed
	 * request but will include an ACK'ed request(s).
	 */
	ack_psn = psn;
	if (aeth >> 29)
		ack_psn--;
	wqe = rvt_get_swqe_ptr(qp, qp->s_acked);
	ibp = to_iport(qp->ibqp.device, qp->port_num);

	/*
	 * The MSN might be for a later WQE than the PSN indicates so
	 * only complete WQEs that the PSN finishes.
	 */
	while ((diff = delta_psn(ack_psn, wqe->lpsn)) >= 0) {
		/*
		 * RDMA_READ_RESPONSE_ONLY is a special case since
		 * we want to generate completion events for everything
		 * before the RDMA read, copy the data, then generate
		 * the completion for the read.
		 */
		if (wqe->wr.opcode == IB_WR_RDMA_READ &&
		    opcode == OP(RDMA_READ_RESPONSE_ONLY) &&
		    diff == 0) {
			ret = 1;
			goto bail_stop;
		}
		/*
		 * If this request is a RDMA read or atomic, and the ACK is
		 * for a later operation, this ACK NAKs the RDMA read or
		 * atomic.  In other words, only a RDMA_READ_LAST or ONLY
		 * can ACK a RDMA read and likewise for atomic ops.  Note
		 * that the NAK case can only happen if relaxed ordering is
		 * used and requests are sent after an RDMA read or atomic
		 * is sent but before the response is received.
		 */
		if ((wqe->wr.opcode == IB_WR_RDMA_READ &&
		     (opcode != OP(RDMA_READ_RESPONSE_LAST) || diff != 0)) ||
		    ((wqe->wr.opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
		      wqe->wr.opcode == IB_WR_ATOMIC_FETCH_AND_ADD) &&
		     (opcode != OP(ATOMIC_ACKNOWLEDGE) || diff != 0))) {
			/* Retry this request. */
			if (!(qp->r_flags & RVT_R_RDMAR_SEQ)) {
				qp->r_flags |= RVT_R_RDMAR_SEQ;
				restart_rc(qp, qp->s_last_psn + 1, 0);
				if (list_empty(&qp->rspwait)) {
					qp->r_flags |= RVT_R_RSP_SEND;
					atomic_inc(&qp->refcount);
					list_add_tail(&qp->rspwait,
						      &rcd->qp_wait_list);
				}
			}
			/*
			 * No need to process the ACK/NAK since we are
			 * restarting an earlier request.
			 */
			goto bail_stop;
		}
		if (wqe->wr.opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
		    wqe->wr.opcode == IB_WR_ATOMIC_FETCH_AND_ADD) {
			u64 *vaddr = wqe->sg_list[0].vaddr;
			*vaddr = val;
		}
		if (qp->s_num_rd_atomic &&
		    (wqe->wr.opcode == IB_WR_RDMA_READ ||
		     wqe->wr.opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
		     wqe->wr.opcode == IB_WR_ATOMIC_FETCH_AND_ADD)) {
			qp->s_num_rd_atomic--;
			/* Restart sending task if fence is complete */
			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);
			} else if (qp->s_flags & RVT_S_WAIT_RDMAR) {
				qp->s_flags &= ~(RVT_S_WAIT_RDMAR |
						 RVT_S_WAIT_ACK);
				hfi1_schedule_send(qp);
			}
		}
		wqe = do_rc_completion(qp, wqe, ibp);
		if (qp->s_acked == qp->s_tail)
			break;
	}

	switch (aeth >> 29) {
	case 0:         /* ACK */
		this_cpu_inc(*ibp->rvp.rc_acks);
		if (qp->s_acked != qp->s_tail) {
			/*
			 * We are expecting more ACKs so
			 * mod the retry timer.
			 */
			hfi1_mod_retry_timer(qp);
			/*
			 * We can stop re-sending the earlier packets and
			 * continue with the next packet the receiver wants.
			 */
			if (cmp_psn(qp->s_psn, psn) <= 0)
				reset_psn(qp, psn + 1);
		} else {
			/* No more acks - kill all timers */
			hfi1_stop_rc_timers(qp);
			if (cmp_psn(qp->s_psn, psn) <= 0) {
				qp->s_state = OP(SEND_LAST);
				qp->s_psn = psn + 1;
			}
		}
		if (qp->s_flags & RVT_S_WAIT_ACK) {
			qp->s_flags &= ~RVT_S_WAIT_ACK;
			hfi1_schedule_send(qp);
		}
		hfi1_get_credit(qp, aeth);
		qp->s_rnr_retry = qp->s_rnr_retry_cnt;
		qp->s_retry = qp->s_retry_cnt;
		update_last_psn(qp, psn);
		return 1;

	case 1:         /* RNR NAK */
		ibp->rvp.n_rnr_naks++;
		if (qp->s_acked == qp->s_tail)
			goto bail_stop;
		if (qp->s_flags & RVT_S_WAIT_RNR)
			goto bail_stop;
		if (qp->s_rnr_retry == 0) {
			status = IB_WC_RNR_RETRY_EXC_ERR;
			goto class_b;
		}
		if (qp->s_rnr_retry_cnt < 7)
			qp->s_rnr_retry--;

		/* The last valid PSN is the previous PSN. */
		update_last_psn(qp, psn - 1);

		ibp->rvp.n_rc_resends += delta_psn(qp->s_psn, psn);

		reset_psn(qp, psn);

		qp->s_flags &= ~(RVT_S_WAIT_SSN_CREDIT | RVT_S_WAIT_ACK);
		hfi1_stop_rc_timers(qp);
		to =
			ib_hfi1_rnr_table[(aeth >> HFI1_AETH_CREDIT_SHIFT) &
					   HFI1_AETH_CREDIT_MASK];
		hfi1_add_rnr_timer(qp, to);
		return 0;

	case 3:         /* NAK */
		if (qp->s_acked == qp->s_tail)
			goto bail_stop;
		/* The last valid PSN is the previous PSN. */
		update_last_psn(qp, psn - 1);
		switch ((aeth >> HFI1_AETH_CREDIT_SHIFT) &
			HFI1_AETH_CREDIT_MASK) {
		case 0: /* PSN sequence error */
			ibp->rvp.n_seq_naks++;
			/*
			 * Back up to the responder's expected PSN.
			 * Note that we might get a NAK in the middle of an
			 * RDMA READ response which terminates the RDMA
			 * READ.
			 */
			restart_rc(qp, psn, 0);
			hfi1_schedule_send(qp);
			break;

		case 1: /* Invalid Request */
			status = IB_WC_REM_INV_REQ_ERR;
			ibp->rvp.n_other_naks++;
			goto class_b;

		case 2: /* Remote Access Error */
			status = IB_WC_REM_ACCESS_ERR;
			ibp->rvp.n_other_naks++;
			goto class_b;

		case 3: /* Remote Operation Error */
			status = IB_WC_REM_OP_ERR;
			ibp->rvp.n_other_naks++;
class_b:
			if (qp->s_last == qp->s_acked) {
				hfi1_send_complete(qp, wqe, status);
				rvt_error_qp(qp, IB_WC_WR_FLUSH_ERR);
			}
			break;

		default:
			/* Ignore other reserved NAK error codes */
			goto reserved;
		}
		qp->s_retry = qp->s_retry_cnt;
		qp->s_rnr_retry = qp->s_rnr_retry_cnt;
		goto bail_stop;

	default:                /* 2: reserved */
reserved:
		/* Ignore reserved NAK codes. */
		goto bail_stop;
	}
	/* cannot be reached  */
bail_stop:
	hfi1_stop_rc_timers(qp);
	return ret;
}

/*
 * We have seen an out of sequence RDMA read middle or last packet.
 * This ACKs SENDs and RDMA writes up to the first RDMA read or atomic SWQE.
 */
static void rdma_seq_err(struct rvt_qp *qp, struct hfi1_ibport *ibp, u32 psn,
			 struct hfi1_ctxtdata *rcd)
{
	struct rvt_swqe *wqe;

	/* Remove QP from retry timer */
	hfi1_stop_rc_timers(qp);

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

	while (cmp_psn(psn, wqe->lpsn) > 0) {
		if (wqe->wr.opcode == IB_WR_RDMA_READ ||
		    wqe->wr.opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
		    wqe->wr.opcode == IB_WR_ATOMIC_FETCH_AND_ADD)
			break;
		wqe = do_rc_completion(qp, wqe, ibp);
	}

	ibp->rvp.n_rdma_seq++;
	qp->r_flags |= RVT_R_RDMAR_SEQ;
	restart_rc(qp, qp->s_last_psn + 1, 0);
	if (list_empty(&qp->rspwait)) {
		qp->r_flags |= RVT_R_RSP_SEND;
		atomic_inc(&qp->refcount);
		list_add_tail(&qp->rspwait, &rcd->qp_wait_list);
	}
}

/**
 * rc_rcv_resp - process an incoming RC response packet
 * @ibp: the port this packet came in on
 * @ohdr: the other headers for this packet
 * @data: the packet data
 * @tlen: the packet length
 * @qp: the QP for this packet
 * @opcode: the opcode for this packet
 * @psn: the packet sequence number for this packet
 * @hdrsize: the header length
 * @pmtu: the path MTU
 *
 * This is called from hfi1_rc_rcv() to process an incoming RC response
 * packet for the given QP.
 * Called at interrupt level.
 */
static void rc_rcv_resp(struct hfi1_ibport *ibp,
			struct hfi1_other_headers *ohdr,
			void *data, u32 tlen, struct rvt_qp *qp,
			u32 opcode, u32 psn, u32 hdrsize, u32 pmtu,
			struct hfi1_ctxtdata *rcd)
{
	struct rvt_swqe *wqe;
	enum ib_wc_status status;
	unsigned long flags;
	int diff;
	u32 pad;
	u32 aeth;
	u64 val;

	spin_lock_irqsave(&qp->s_lock, flags);

	trace_hfi1_ack(qp, psn);

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

	/* Ignore duplicate responses. */
	diff = cmp_psn(psn, qp->s_last_psn);
	if (unlikely(diff <= 0)) {
		/* Update credits for "ghost" ACKs */
		if (diff == 0 && opcode == OP(ACKNOWLEDGE)) {
			aeth = be32_to_cpu(ohdr->u.aeth);
			if ((aeth >> 29) == 0)
				hfi1_get_credit(qp, aeth);
		}
		goto ack_done;
	}

	/*
	 * Skip everything other than the PSN we expect, if we are waiting
	 * for a reply to a restarted RDMA read or atomic op.
	 */
	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;
	}

	if (unlikely(qp->s_acked == qp->s_tail))
		goto ack_done;
	wqe = rvt_get_swqe_ptr(qp, qp->s_acked);
	status = IB_WC_SUCCESS;

	switch (opcode) {
	case OP(ACKNOWLEDGE):
	case OP(ATOMIC_ACKNOWLEDGE):
	case OP(RDMA_READ_RESPONSE_FIRST):
		aeth = be32_to_cpu(ohdr->u.aeth);
		if (opcode == OP(ATOMIC_ACKNOWLEDGE)) {
			__be32 *p = ohdr->u.at.atomic_ack_eth;

			val = ((u64)be32_to_cpu(p[0]) << 32) |
				be32_to_cpu(p[1]);
		} else {
			val = 0;
		}
		if (!do_rc_ack(qp, aeth, psn, opcode, val, rcd) ||
		    opcode != OP(RDMA_READ_RESPONSE_FIRST))
			goto ack_done;
		wqe = rvt_get_swqe_ptr(qp, qp->s_acked);
		if (unlikely(wqe->wr.opcode != IB_WR_RDMA_READ))
			goto ack_op_err;
		/*
		 * If this is a response to a resent RDMA read, we
		 * have to be careful to copy the data to the right
		 * location.
		 */
		qp->s_rdma_read_len = restart_sge(&qp->s_rdma_read_sge,
						  wqe, psn, pmtu);
		goto read_middle;

	case OP(RDMA_READ_RESPONSE_MIDDLE):
		/* no AETH, no ACK */
		if (unlikely(cmp_psn(psn, qp->s_last_psn + 1)))
			goto ack_seq_err;
		if (unlikely(wqe->wr.opcode != IB_WR_RDMA_READ))
			goto ack_op_err;
read_middle:
		if (unlikely(tlen != (hdrsize + pmtu + 4)))
			goto ack_len_err;
		if (unlikely(pmtu >= qp->s_rdma_read_len))
			goto ack_len_err;

		/*
		 * We got a response so update the timeout.
		 * 4.096 usec. * (1 << qp->timeout)
		 */
		qp->s_flags |= RVT_S_TIMER;
		mod_timer(&qp->s_timer, jiffies + qp->timeout_jiffies);
		if (qp->s_flags & RVT_S_WAIT_ACK) {
			qp->s_flags &= ~RVT_S_WAIT_ACK;
			hfi1_schedule_send(qp);
		}

		if (opcode == OP(RDMA_READ_RESPONSE_MIDDLE))
			qp->s_retry = qp->s_retry_cnt;

		/*
		 * Update the RDMA receive state but do the copy w/o
		 * holding the locks and blocking interrupts.
		 */
		qp->s_rdma_read_len -= pmtu;
		update_last_psn(qp, psn);
		spin_unlock_irqrestore(&qp->s_lock, flags);
		hfi1_copy_sge(&qp->s_rdma_read_sge, data, pmtu, 0, 0);
		goto bail;

	case OP(RDMA_READ_RESPONSE_ONLY):
		aeth = be32_to_cpu(ohdr->u.aeth);
		if (!do_rc_ack(qp, aeth, psn, opcode, 0, rcd))
			goto ack_done;
		/* Get the number of bytes the message was padded by. */
		pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
		/*
		 * Check that the data size is >= 0 && <= pmtu.
		 * Remember to account for ICRC (4).
		 */
		if (unlikely(tlen < (hdrsize + pad + 4)))
			goto ack_len_err;
		/*
		 * If this is a response to a resent RDMA read, we
		 * have to be careful to copy the data to the right
		 * location.
		 */
		wqe = rvt_get_swqe_ptr(qp, qp->s_acked);
		qp->s_rdma_read_len = restart_sge(&qp->s_rdma_read_sge,
						  wqe, psn, pmtu);
		goto read_last;

	case OP(RDMA_READ_RESPONSE_LAST):
		/* ACKs READ req. */
		if (unlikely(cmp_psn(psn, qp->s_last_psn + 1)))
			goto ack_seq_err;
		if (unlikely(wqe->wr.opcode != IB_WR_RDMA_READ))
			goto ack_op_err;
		/* Get the number of bytes the message was padded by. */
		pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
		/*
		 * Check that the data size is >= 1 && <= pmtu.
		 * Remember to account for ICRC (4).
		 */
		if (unlikely(tlen <= (hdrsize + pad + 4)))
			goto ack_len_err;
read_last:
		tlen -= hdrsize + pad + 4;
		if (unlikely(tlen != qp->s_rdma_read_len))
			goto ack_len_err;
		aeth = be32_to_cpu(ohdr->u.aeth);
		hfi1_copy_sge(&qp->s_rdma_read_sge, data, tlen, 0, 0);
		WARN_ON(qp->s_rdma_read_sge.num_sge);
		(void)do_rc_ack(qp, aeth, psn,
				 OP(RDMA_READ_RESPONSE_LAST), 0, rcd);
		goto ack_done;
	}

ack_op_err:
	status = IB_WC_LOC_QP_OP_ERR;
	goto ack_err;

ack_seq_err:
	rdma_seq_err(qp, ibp, psn, rcd);
	goto ack_done;

ack_len_err:
	status = IB_WC_LOC_LEN_ERR;
ack_err:
	if (qp->s_last == qp->s_acked) {
		hfi1_send_complete(qp, wqe, status);
		rvt_error_qp(qp, IB_WC_WR_FLUSH_ERR);
	}
ack_done:
	spin_unlock_irqrestore(&qp->s_lock, flags);
bail:
	return;
}

static inline void rc_defered_ack(struct hfi1_ctxtdata *rcd,
				  struct rvt_qp *qp)
{
	if (list_empty(&qp->rspwait)) {
		qp->r_flags |= RVT_R_RSP_NAK;
		atomic_inc(&qp->refcount);
		list_add_tail(&qp->rspwait, &rcd->qp_wait_list);
	}
}

static inline void rc_cancel_ack(struct rvt_qp *qp)
{
	struct hfi1_qp_priv *priv = qp->priv;

	priv->r_adefered = 0;
	if (list_empty(&qp->rspwait))
		return;
	list_del_init(&qp->rspwait);
	qp->r_flags &= ~RVT_R_RSP_NAK;
	if (atomic_dec_and_test(&qp->refcount))
		wake_up(&qp->wait);
}

/**
 * rc_rcv_error - process an incoming duplicate or error RC packet
 * @ohdr: the other headers for this packet
 * @data: the packet data
 * @qp: the QP for this packet
 * @opcode: the opcode for this packet
 * @psn: the packet sequence number for this packet
 * @diff: the difference between the PSN and the expected PSN
 *
 * This is called from hfi1_rc_rcv() to process an unexpected
 * incoming RC packet for the given QP.
 * Called at interrupt level.
 * Return 1 if no more processing is needed; otherwise return 0 to
 * schedule a response to be sent.
 */
static noinline int rc_rcv_error(struct hfi1_other_headers *ohdr, void *data,
				 struct rvt_qp *qp, u32 opcode, u32 psn,
				 int diff, struct hfi1_ctxtdata *rcd)
{
	struct hfi1_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
	struct rvt_ack_entry *e;
	unsigned long flags;
	u8 i, prev;
	int old_req;

	trace_hfi1_rcv_error(qp, psn);
	if (diff > 0) {
		/*
		 * Packet sequence error.
		 * A NAK will ACK earlier sends and RDMA writes.
		 * Don't queue the NAK if we already sent one.
		 */
		if (!qp->r_nak_state) {
			ibp->rvp.n_rc_seqnak++;
			qp->r_nak_state = IB_NAK_PSN_ERROR;
			/* Use the expected PSN. */
			qp->r_ack_psn = qp->r_psn;
			/*
			 * Wait to send the sequence NAK until all packets
			 * in the receive queue have been processed.
			 * Otherwise, we end up propagating congestion.
			 */
			rc_defered_ack(rcd, qp);
		}
		goto done;
	}

	/*
	 * Handle a duplicate request.  Don't re-execute SEND, RDMA
	 * write or atomic op.  Don't NAK errors, just silently drop
	 * the duplicate request.  Note that r_sge, r_len, and
	 * r_rcv_len may be in use so don't modify them.
	 *
	 * We are supposed to ACK the earliest duplicate PSN but we
	 * can coalesce an outstanding duplicate ACK.  We have to
	 * send the earliest so that RDMA reads can be restarted at
	 * the requester's expected PSN.
	 *
	 * First, find where this duplicate PSN falls within the
	 * ACKs previously sent.
	 * old_req is true if there is an older response that is scheduled
	 * to be sent before sending this one.
	 */
	e = NULL;
	old_req = 1;
	ibp->rvp.n_rc_dupreq++;

	spin_lock_irqsave(&qp->s_lock, flags);

	for (i = qp->r_head_ack_queue; ; i = prev) {
		if (i == qp->s_tail_ack_queue)
			old_req = 0;
		if (i)
			prev = i - 1;
		else
			prev = HFI1_MAX_RDMA_ATOMIC;
		if (prev == qp->r_head_ack_queue) {
			e = NULL;
			break;
		}
		e = &qp->s_ack_queue[prev];
		if (!e->opcode) {
			e = NULL;
			break;
		}
		if (cmp_psn(psn, e->psn) >= 0) {
			if (prev == qp->s_tail_ack_queue &&
			    cmp_psn(psn, e->lpsn) <= 0)
				old_req = 0;
			break;
		}
	}
	switch (opcode) {
	case OP(RDMA_READ_REQUEST): {
		struct ib_reth *reth;
		u32 offset;
		u32 len;

		/*
		 * If we didn't find the RDMA read request in the ack queue,
		 * we can ignore this request.
		 */
		if (!e || e->opcode != OP(RDMA_READ_REQUEST))
			goto unlock_done;
		/* RETH comes after BTH */
		reth = &ohdr->u.rc.reth;
		/*
		 * Address range must be a subset of the original
		 * request and start on pmtu boundaries.
		 * We reuse the old ack_queue slot since the requester
		 * should not back up and request an earlier PSN for the
		 * same request.
		 */
		offset = delta_psn(psn, e->psn) * qp->pmtu;
		len = be32_to_cpu(reth->length);
		if (unlikely(offset + len != e->rdma_sge.sge_length))
			goto unlock_done;
		if (e->rdma_sge.mr) {
			rvt_put_mr(e->rdma_sge.mr);
			e->rdma_sge.mr = NULL;
		}
		if (len != 0) {
			u32 rkey = be32_to_cpu(reth->rkey);
			u64 vaddr = be64_to_cpu(reth->vaddr);
			int ok;

			ok = rvt_rkey_ok(qp, &e->rdma_sge, len, vaddr, rkey,
					 IB_ACCESS_REMOTE_READ);
			if (unlikely(!ok))
				goto unlock_done;
		} else {
			e->rdma_sge.vaddr = NULL;
			e->rdma_sge.length = 0;
			e->rdma_sge.sge_length = 0;
		}
		e->psn = psn;
		if (old_req)
			goto unlock_done;
		qp->s_tail_ack_queue = prev;
		break;
	}

	case OP(COMPARE_SWAP):
	case OP(FETCH_ADD): {
		/*
		 * If we didn't find the atomic request in the ack queue
		 * or the send tasklet is already backed up to send an
		 * earlier entry, we can ignore this request.
		 */
		if (!e || e->opcode != (u8)opcode || old_req)
			goto unlock_done;
		qp->s_tail_ack_queue = prev;
		break;
	}

	default:
		/*
		 * Ignore this operation if it doesn't request an ACK
		 * or an earlier RDMA read or atomic is going to be resent.
		 */
		if (!(psn & IB_BTH_REQ_ACK) || old_req)
			goto unlock_done;
		/*
		 * Resend the most recent ACK if this request is
		 * after all the previous RDMA reads and atomics.
		 */
		if (i == qp->r_head_ack_queue) {
			spin_unlock_irqrestore(&qp->s_lock, flags);
			qp->r_nak_state = 0;
			qp->r_ack_psn = qp->r_psn - 1;
			goto send_ack;
		}

		/*
		 * Resend the RDMA read or atomic op which
		 * ACKs this duplicate request.
		 */
		qp->s_tail_ack_queue = i;
		break;
	}
	qp->s_ack_state = OP(ACKNOWLEDGE);
	qp->s_flags |= RVT_S_RESP_PENDING;
	qp->r_nak_state = 0;
	hfi1_schedule_send(qp);

unlock_done:
	spin_unlock_irqrestore(&qp->s_lock, flags);
done:
	return 1;

send_ack:
	return 0;
}

void hfi1_rc_error(struct rvt_qp *qp, enum ib_wc_status err)
{
	unsigned long flags;
	int lastwqe;

	spin_lock_irqsave(&qp->s_lock, flags);
	lastwqe = rvt_error_qp(qp, err);
	spin_unlock_irqrestore(&qp->s_lock, flags);

	if (lastwqe) {
		struct ib_event ev;

		ev.device = qp->ibqp.device;
		ev.element.qp = &qp->ibqp;
		ev.event = IB_EVENT_QP_LAST_WQE_REACHED;
		qp->ibqp.event_handler(&ev, qp->ibqp.qp_context);
	}
}

static inline void update_ack_queue(struct rvt_qp *qp, unsigned n)
{
	unsigned next;

	next = n + 1;
	if (next > HFI1_MAX_RDMA_ATOMIC)
		next = 0;
	qp->s_tail_ack_queue = next;
	qp->s_ack_state = OP(ACKNOWLEDGE);
}

static void log_cca_event(struct hfi1_pportdata *ppd, u8 sl, u32 rlid,
			  u32 lqpn, u32 rqpn, u8 svc_type)
{
	struct opa_hfi1_cong_log_event_internal *cc_event;
	unsigned long flags;

	if (sl >= OPA_MAX_SLS)
		return;

	spin_lock_irqsave(&ppd->cc_log_lock, flags);

	ppd->threshold_cong_event_map[sl / 8] |= 1 << (sl % 8);
	ppd->threshold_event_counter++;

	cc_event = &ppd->cc_events[ppd->cc_log_idx++];
	if (ppd->cc_log_idx == OPA_CONG_LOG_ELEMS)
		ppd->cc_log_idx = 0;
	cc_event->lqpn = lqpn & RVT_QPN_MASK;
	cc_event->rqpn = rqpn & RVT_QPN_MASK;
	cc_event->sl = sl;
	cc_event->svc_type = svc_type;
	cc_event->rlid = rlid;
	/* keep timestamp in units of 1.024 usec */
	cc_event->timestamp = ktime_to_ns(ktime_get()) / 1024;

	spin_unlock_irqrestore(&ppd->cc_log_lock, flags);
}

void process_becn(struct hfi1_pportdata *ppd, u8 sl, u16 rlid, u32 lqpn,
		  u32 rqpn, u8 svc_type)
{
	struct cca_timer *cca_timer;
	u16 ccti, ccti_incr, ccti_timer, ccti_limit;
	u8 trigger_threshold;
	struct cc_state *cc_state;
	unsigned long flags;

	if (sl >= OPA_MAX_SLS)
		return;

	cc_state = get_cc_state(ppd);

	if (!cc_state)
		return;

	/*
	 * 1) increase CCTI (for this SL)
	 * 2) select IPG (i.e., call set_link_ipg())
	 * 3) start timer
	 */
	ccti_limit = cc_state->cct.ccti_limit;
	ccti_incr = cc_state->cong_setting.entries[sl].ccti_increase;
	ccti_timer = cc_state->cong_setting.entries[sl].ccti_timer;
	trigger_threshold =
		cc_state->cong_setting.entries[sl].trigger_threshold;

	spin_lock_irqsave(&ppd->cca_timer_lock, flags);

	cca_timer = &ppd->cca_timer[sl];
	if (cca_timer->ccti < ccti_limit) {
		if (cca_timer->ccti + ccti_incr <= ccti_limit)
			cca_timer->ccti += ccti_incr;
		else
			cca_timer->ccti = ccti_limit;
		set_link_ipg(ppd);
	}

	ccti = cca_timer->ccti;

	if (!hrtimer_active(&cca_timer->hrtimer)) {
		/* ccti_timer is in units of 1.024 usec */
		unsigned long nsec = 1024 * ccti_timer;

		hrtimer_start(&cca_timer->hrtimer, ns_to_ktime(nsec),
			      HRTIMER_MODE_REL);
	}

	spin_unlock_irqrestore(&ppd->cca_timer_lock, flags);

	if ((trigger_threshold != 0) && (ccti >= trigger_threshold))
		log_cca_event(ppd, sl, rlid, lqpn, rqpn, svc_type);
}

/**
 * hfi1_rc_rcv - process an incoming RC packet
 * @rcd: the context pointer
 * @hdr: the header of this packet
 * @rcv_flags: flags relevant to rcv processing
 * @data: the packet data
 * @tlen: the packet length
 * @qp: the QP for this packet
 *
 * This is called from qp_rcv() to process an incoming RC packet
 * for the given QP.
 * May be called at interrupt level.
 */
void hfi1_rc_rcv(struct hfi1_packet *packet)
{
	struct hfi1_ctxtdata *rcd = packet->rcd;
	struct hfi1_ib_header *hdr = packet->hdr;
	u32 rcv_flags = packet->rcv_flags;
	void *data = packet->ebuf;
	u32 tlen = packet->tlen;
	struct rvt_qp *qp = packet->qp;
	struct hfi1_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
	struct hfi1_other_headers *ohdr = packet->ohdr;
	u32 bth0, opcode;
	u32 hdrsize = packet->hlen;
	u32 psn;
	u32 pad;
	struct ib_wc wc;
	u32 pmtu = qp->pmtu;
	int diff;
	struct ib_reth *reth;
	unsigned long flags;
	int ret, is_fecn = 0;
	int copy_last = 0;
	u32 rkey;

	bth0 = be32_to_cpu(ohdr->bth[0]);
	if (hfi1_ruc_check_hdr(ibp, hdr, rcv_flags & HFI1_HAS_GRH, qp, bth0))
		return;

	is_fecn = process_ecn(qp, packet, false);

	psn = be32_to_cpu(ohdr->bth[2]);
	opcode = (bth0 >> 24) & 0xff;

	/*
	 * Process responses (ACKs) before anything else.  Note that the
	 * packet sequence number will be for something in the send work
	 * queue rather than the expected receive packet sequence number.
	 * In other words, this QP is the requester.
	 */
	if (opcode >= OP(RDMA_READ_RESPONSE_FIRST) &&
	    opcode <= OP(ATOMIC_ACKNOWLEDGE)) {
		rc_rcv_resp(ibp, ohdr, data, tlen, qp, opcode, psn,
			    hdrsize, pmtu, rcd);
		if (is_fecn)
			goto send_ack;
		return;
	}

	/* Compute 24 bits worth of difference. */
	diff = delta_psn(psn, qp->r_psn);
	if (unlikely(diff)) {
		if (rc_rcv_error(ohdr, data, qp, opcode, psn, diff, rcd))
			return;
		goto send_ack;
	}

	/* Check for opcode sequence errors. */
	switch (qp->r_state) {
	case OP(SEND_FIRST):
	case OP(SEND_MIDDLE):
		if (opcode == OP(SEND_MIDDLE) ||
		    opcode == OP(SEND_LAST) ||
		    opcode == OP(SEND_LAST_WITH_IMMEDIATE) ||
		    opcode == OP(SEND_LAST_WITH_INVALIDATE))
			break;
		goto nack_inv;

	case OP(RDMA_WRITE_FIRST):
	case OP(RDMA_WRITE_MIDDLE):
		if (opcode == OP(RDMA_WRITE_MIDDLE) ||
		    opcode == OP(RDMA_WRITE_LAST) ||
		    opcode == OP(RDMA_WRITE_LAST_WITH_IMMEDIATE))
			break;
		goto nack_inv;

	default:
		if (opcode == OP(SEND_MIDDLE) ||
		    opcode == OP(SEND_LAST) ||
		    opcode == OP(SEND_LAST_WITH_IMMEDIATE) ||
		    opcode == OP(SEND_LAST_WITH_INVALIDATE) ||
		    opcode == OP(RDMA_WRITE_MIDDLE) ||
		    opcode == OP(RDMA_WRITE_LAST) ||
		    opcode == OP(RDMA_WRITE_LAST_WITH_IMMEDIATE))
			goto nack_inv;
		/*
		 * Note that it is up to the requester to not send a new
		 * RDMA read or atomic operation before receiving an ACK
		 * for the previous operation.
		 */
		break;
	}

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

	/* OK, process the packet. */
	switch (opcode) {
	case OP(SEND_FIRST):
		ret = hfi1_rvt_get_rwqe(qp, 0);
		if (ret < 0)
			goto nack_op_err;
		if (!ret)
			goto rnr_nak;
		qp->r_rcv_len = 0;
		/* FALLTHROUGH */
	case OP(SEND_MIDDLE):
	case OP(RDMA_WRITE_MIDDLE):
send_middle:
		/* Check for invalid length PMTU or posted rwqe len. */
		if (unlikely(tlen != (hdrsize + pmtu + 4)))
			goto nack_inv;
		qp->r_rcv_len += pmtu;
		if (unlikely(qp->r_rcv_len > qp->r_len))
			goto nack_inv;
		hfi1_copy_sge(&qp->r_sge, data, pmtu, 1, 0);
		break;

	case OP(RDMA_WRITE_LAST_WITH_IMMEDIATE):
		/* consume RWQE */
		ret = hfi1_rvt_get_rwqe(qp, 1);
		if (ret < 0)
			goto nack_op_err;
		if (!ret)
			goto rnr_nak;
		goto send_last_imm;

	case OP(SEND_ONLY):
	case OP(SEND_ONLY_WITH_IMMEDIATE):
	case OP(SEND_ONLY_WITH_INVALIDATE):
		ret = hfi1_rvt_get_rwqe(qp, 0);
		if (ret < 0)
			goto nack_op_err;
		if (!ret)
			goto rnr_nak;
		qp->r_rcv_len = 0;
		if (opcode == OP(SEND_ONLY))
			goto no_immediate_data;
		if (opcode == OP(SEND_ONLY_WITH_INVALIDATE))
			goto send_last_inv;
		/* FALLTHROUGH for SEND_ONLY_WITH_IMMEDIATE */
	case OP(SEND_LAST_WITH_IMMEDIATE):
send_last_imm:
		wc.ex.imm_data = ohdr->u.imm_data;
		wc.wc_flags = IB_WC_WITH_IMM;
		goto send_last;
	case OP(SEND_LAST_WITH_INVALIDATE):
send_last_inv:
		rkey = be32_to_cpu(ohdr->u.ieth);
		if (rvt_invalidate_rkey(qp, rkey))
			goto no_immediate_data;
		wc.ex.invalidate_rkey = rkey;
		wc.wc_flags = IB_WC_WITH_INVALIDATE;
		goto send_last;
	case OP(RDMA_WRITE_LAST):
		copy_last = ibpd_to_rvtpd(qp->ibqp.pd)->user;
		/* fall through */
	case OP(SEND_LAST):
no_immediate_data:
		wc.wc_flags = 0;
		wc.ex.imm_data = 0;
send_last:
		/* Get the number of bytes the message was padded by. */
		pad = (bth0 >> 20) & 3;
		/* Check for invalid length. */
		/* LAST len should be >= 1 */
		if (unlikely(tlen < (hdrsize + pad + 4)))
			goto nack_inv;
		/* Don't count the CRC. */
		tlen -= (hdrsize + pad + 4);
		wc.byte_len = tlen + qp->r_rcv_len;
		if (unlikely(wc.byte_len > qp->r_len))
			goto nack_inv;
		hfi1_copy_sge(&qp->r_sge, data, tlen, 1, copy_last);
		rvt_put_ss(&qp->r_sge);
		qp->r_msn++;
		if (!test_and_clear_bit(RVT_R_WRID_VALID, &qp->r_aflags))
			break;
		wc.wr_id = qp->r_wr_id;
		wc.status = IB_WC_SUCCESS;
		if (opcode == OP(RDMA_WRITE_LAST_WITH_IMMEDIATE) ||
		    opcode == OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE))
			wc.opcode = IB_WC_RECV_RDMA_WITH_IMM;
		else
			wc.opcode = IB_WC_RECV;
		wc.qp = &qp->ibqp;
		wc.src_qp = qp->remote_qpn;
		wc.slid = qp->remote_ah_attr.dlid;
		/*
		 * It seems that IB mandates the presence of an SL in a
		 * work completion only for the UD transport (see section
		 * 11.4.2 of IBTA Vol. 1).
		 *
		 * However, the way the SL is chosen below is consistent
		 * with the way that IB/qib works and is trying avoid
		 * introducing incompatibilities.
		 *
		 * See also OPA Vol. 1, section 9.7.6, and table 9-17.
		 */
		wc.sl = qp->remote_ah_attr.sl;
		/* zero fields that are N/A */
		wc.vendor_err = 0;
		wc.pkey_index = 0;
		wc.dlid_path_bits = 0;
		wc.port_num = 0;
		/* Signal completion event if the solicited bit is set. */
		rvt_cq_enter(ibcq_to_rvtcq(qp->ibqp.recv_cq), &wc,
			     (bth0 & IB_BTH_SOLICITED) != 0);
		break;

	case OP(RDMA_WRITE_ONLY):
		copy_last = 1;
		/* fall through */
	case OP(RDMA_WRITE_FIRST):
	case OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE):
		if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_WRITE)))
			goto nack_inv;
		/* consume RWQE */
		reth = &ohdr->u.rc.reth;
		qp->r_len = be32_to_cpu(reth->length);
		qp->r_rcv_len = 0;
		qp->r_sge.sg_list = NULL;
		if (qp->r_len != 0) {
			u32 rkey = be32_to_cpu(reth->rkey);
			u64 vaddr = be64_to_cpu(reth->vaddr);
			int ok;

			/* Check rkey & NAK */
			ok = rvt_rkey_ok(qp, &qp->r_sge.sge, qp->r_len, vaddr,
					 rkey, IB_ACCESS_REMOTE_WRITE);
			if (unlikely(!ok))
				goto nack_acc;
			qp->r_sge.num_sge = 1;
		} else {
			qp->r_sge.num_sge = 0;
			qp->r_sge.sge.mr = NULL;
			qp->r_sge.sge.vaddr = NULL;
			qp->r_sge.sge.length = 0;
			qp->r_sge.sge.sge_length = 0;
		}
		if (opcode == OP(RDMA_WRITE_FIRST))
			goto send_middle;
		else if (opcode == OP(RDMA_WRITE_ONLY))
			goto no_immediate_data;
		ret = hfi1_rvt_get_rwqe(qp, 1);
		if (ret < 0)
			goto nack_op_err;
		if (!ret)
			goto rnr_nak;
		wc.ex.imm_data = ohdr->u.rc.imm_data;
		wc.wc_flags = IB_WC_WITH_IMM;
		goto send_last;

	case OP(RDMA_READ_REQUEST): {
		struct rvt_ack_entry *e;
		u32 len;
		u8 next;

		if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_READ)))
			goto nack_inv;
		next = qp->r_head_ack_queue + 1;
		/* s_ack_queue is size HFI1_MAX_RDMA_ATOMIC+1 so use > not >= */
		if (next > HFI1_MAX_RDMA_ATOMIC)
			next = 0;
		spin_lock_irqsave(&qp->s_lock, flags);
		if (unlikely(next == qp->s_tail_ack_queue)) {
			if (!qp->s_ack_queue[next].sent)
				goto nack_inv_unlck;
			update_ack_queue(qp, next);
		}
		e = &qp->s_ack_queue[qp->r_head_ack_queue];
		if (e->opcode == OP(RDMA_READ_REQUEST) && e->rdma_sge.mr) {
			rvt_put_mr(e->rdma_sge.mr);
			e->rdma_sge.mr = NULL;
		}
		reth = &ohdr->u.rc.reth;
		len = be32_to_cpu(reth->length);
		if (len) {
			u32 rkey = be32_to_cpu(reth->rkey);
			u64 vaddr = be64_to_cpu(reth->vaddr);
			int ok;

			/* Check rkey & NAK */
			ok = rvt_rkey_ok(qp, &e->rdma_sge, len, vaddr,
					 rkey, IB_ACCESS_REMOTE_READ);
			if (unlikely(!ok))
				goto nack_acc_unlck;
			/*
			 * Update the next expected PSN.  We add 1 later
			 * below, so only add the remainder here.
			 */
			if (len > pmtu)
				qp->r_psn += (len - 1) / pmtu;
		} else {
			e->rdma_sge.mr = NULL;
			e->rdma_sge.vaddr = NULL;
			e->rdma_sge.length = 0;
			e->rdma_sge.sge_length = 0;
		}
		e->opcode = opcode;
		e->sent = 0;
		e->psn = psn;
		e->lpsn = qp->r_psn;
		/*
		 * 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++;
		qp->r_state = opcode;
		qp->r_nak_state = 0;
		qp->r_head_ack_queue = next;

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

		spin_unlock_irqrestore(&qp->s_lock, flags);
		if (is_fecn)
			goto send_ack;
		return;
	}

	case OP(COMPARE_SWAP):
	case OP(FETCH_ADD): {
		struct ib_atomic_eth *ateth;
		struct rvt_ack_entry *e;
		u64 vaddr;
		atomic64_t *maddr;
		u64 sdata;
		u32 rkey;
		u8 next;

		if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_ATOMIC)))
			goto nack_inv;
		next = qp->r_head_ack_queue + 1;
		if (next > HFI1_MAX_RDMA_ATOMIC)
			next = 0;
		spin_lock_irqsave(&qp->s_lock, flags);
		if (unlikely(next == qp->s_tail_ack_queue)) {
			if (!qp->s_ack_queue[next].sent)
				goto nack_inv_unlck;
			update_ack_queue(qp, next);
		}
		e = &qp->s_ack_queue[qp->r_head_ack_queue];
		if (e->opcode == OP(RDMA_READ_REQUEST) && e->rdma_sge.mr) {
			rvt_put_mr(e->rdma_sge.mr);
			e->rdma_sge.mr = NULL;
		}
		ateth = &ohdr->u.atomic_eth;
		vaddr = ((u64)be32_to_cpu(ateth->vaddr[0]) << 32) |
			be32_to_cpu(ateth->vaddr[1]);
		if (unlikely(vaddr & (sizeof(u64) - 1)))
			goto nack_inv_unlck;
		rkey = be32_to_cpu(ateth->rkey);
		/* Check rkey & NAK */
		if (unlikely(!rvt_rkey_ok(qp, &qp->r_sge.sge, sizeof(u64),
					  vaddr, rkey,
					  IB_ACCESS_REMOTE_ATOMIC)))
			goto nack_acc_unlck;
		/* Perform atomic OP and save result. */
		maddr = (atomic64_t *)qp->r_sge.sge.vaddr;
		sdata = be64_to_cpu(ateth->swap_data);
		e->atomic_data = (opcode == OP(FETCH_ADD)) ?
			(u64)atomic64_add_return(sdata, maddr) - sdata :
			(u64)cmpxchg((u64 *)qp->r_sge.sge.vaddr,
				      be64_to_cpu(ateth->compare_data),
				      sdata);
		rvt_put_mr(qp->r_sge.sge.mr);
		qp->r_sge.num_sge = 0;
		e->opcode = opcode;
		e->sent = 0;
		e->psn = psn;
		e->lpsn = psn;
		qp->r_msn++;
		qp->r_psn++;
		qp->r_state = opcode;
		qp->r_nak_state = 0;
		qp->r_head_ack_queue = next;

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

		spin_unlock_irqrestore(&qp->s_lock, flags);
		if (is_fecn)
			goto send_ack;
		return;
	}

	default:
		/* NAK unknown opcodes. */
		goto nack_inv;
	}
	qp->r_psn++;
	qp->r_state = opcode;
	qp->r_ack_psn = psn;
	qp->r_nak_state = 0;
	/* Send an ACK if requested or required. */
	if (psn & IB_BTH_REQ_ACK) {
		struct hfi1_qp_priv *priv = qp->priv;

		if (packet->numpkt == 0) {
			rc_cancel_ack(qp);
			goto send_ack;
		}
		if (priv->r_adefered >= HFI1_PSN_CREDIT) {
			rc_cancel_ack(qp);
			goto send_ack;
		}
		if (unlikely(is_fecn)) {
			rc_cancel_ack(qp);
			goto send_ack;
		}
		priv->r_adefered++;
		rc_defered_ack(rcd, qp);
	}
	return;

rnr_nak:
	qp->r_nak_state = qp->r_min_rnr_timer | IB_RNR_NAK;
	qp->r_ack_psn = qp->r_psn;
	/* Queue RNR NAK for later */
	rc_defered_ack(rcd, qp);
	return;

nack_op_err:
	hfi1_rc_error(qp, IB_WC_LOC_QP_OP_ERR);
	qp->r_nak_state = IB_NAK_REMOTE_OPERATIONAL_ERROR;
	qp->r_ack_psn = qp->r_psn;
	/* Queue NAK for later */
	rc_defered_ack(rcd, qp);
	return;

nack_inv_unlck:
	spin_unlock_irqrestore(&qp->s_lock, flags);
nack_inv:
	hfi1_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_unlck:
	spin_unlock_irqrestore(&qp->s_lock, flags);
nack_acc:
	hfi1_rc_error(qp, IB_WC_LOC_PROT_ERR);
	qp->r_nak_state = IB_NAK_REMOTE_ACCESS_ERROR;
	qp->r_ack_psn = qp->r_psn;
send_ack:
	hfi1_send_rc_ack(rcd, qp, is_fecn);
}

void hfi1_rc_hdrerr(
	struct hfi1_ctxtdata *rcd,
	struct hfi1_ib_header *hdr,
	u32 rcv_flags,
	struct rvt_qp *qp)
{
	int has_grh = rcv_flags & HFI1_HAS_GRH;
	struct hfi1_other_headers *ohdr;
	struct hfi1_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
	int diff;
	u32 opcode;
	u32 psn, bth0;

	/* Check for GRH */
	ohdr = &hdr->u.oth;
	if (has_grh)
		ohdr = &hdr->u.l.oth;

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

	psn = be32_to_cpu(ohdr->bth[2]);
	opcode = (bth0 >> 24) & 0xff;

	/* Only deal with RDMA Writes for now */
	if (opcode < IB_OPCODE_RC_RDMA_READ_RESPONSE_FIRST) {
		diff = delta_psn(psn, qp->r_psn);
		if (!qp->r_nak_state && diff >= 0) {
			ibp->rvp.n_rc_seqnak++;
			qp->r_nak_state = IB_NAK_PSN_ERROR;
			/* Use the expected PSN. */
			qp->r_ack_psn = qp->r_psn;
			/*
			 * Wait to send the sequence
			 * NAK until all packets
			 * in the receive queue have
			 * been processed.
			 * Otherwise, we end up
			 * propagating congestion.
			 */
			rc_defered_ack(rcd, qp);
		} /* Out of sequence NAK */
	} /* QP Request NAKs */
}