aboutsummaryrefslogtreecommitdiff
path: root/drivers/infiniband/hw/ipath/ipath_rc.c
blob: 774d1615ce2f134f7058ffb72880190f25c1adde (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
/*
 * Copyright (c) 2006 QLogic, Inc. All rights reserved.
 * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
 *
 * This software is available to you under a choice of one of two
 * licenses.  You may choose to be licensed under the terms of the GNU
 * General Public License (GPL) Version 2, available from the file
 * COPYING in the main directory of this source tree, or the
 * OpenIB.org BSD license below:
 *
 *     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.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 */

#include "ipath_verbs.h"
#include "ipath_common.h"

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

/**
 * ipath_init_restart- initialize the qp->s_sge after a restart
 * @qp: the QP who's SGE we're restarting
 * @wqe: the work queue to initialize the QP's SGE from
 *
 * The QP s_lock should be held and interrupts disabled.
 */
static void ipath_init_restart(struct ipath_qp *qp, struct ipath_swqe *wqe)
{
	struct ipath_ibdev *dev;
	u32 len;

	len = ((qp->s_psn - wqe->psn) & IPATH_PSN_MASK) *
		ib_mtu_enum_to_int(qp->path_mtu);
	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;
	ipath_skip_sge(&qp->s_sge, len);
	qp->s_len = wqe->length - len;
	dev = to_idev(qp->ibqp.device);
	spin_lock(&dev->pending_lock);
	if (list_empty(&qp->timerwait))
		list_add_tail(&qp->timerwait,
			      &dev->pending[dev->pending_index]);
	spin_unlock(&dev->pending_lock);
}

/**
 * ipath_make_rc_ack - construct a response packet (ACK, NAK, or RDMA read)
 * @qp: a pointer to the QP
 * @ohdr: a pointer to the IB header being constructed
 * @pmtu: the path MTU
 *
 * Return bth0 if constructed; otherwise, return 0.
 * Note the QP s_lock must be held.
 */
u32 ipath_make_rc_ack(struct ipath_qp *qp,
		      struct ipath_other_headers *ohdr,
		      u32 pmtu)
{
	u32 hwords;
	u32 len;
	u32 bth0;

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

	/*
	 * Send a response.  Note that we are in the responder's
	 * side of the QP context.
	 */
	switch (qp->s_ack_state) {
	case OP(RDMA_READ_REQUEST):
		qp->s_cur_sge = &qp->s_rdma_sge;
		len = qp->s_rdma_len;
		if (len > pmtu) {
			len = pmtu;
			qp->s_ack_state = OP(RDMA_READ_RESPONSE_FIRST);
		} else
			qp->s_ack_state = OP(RDMA_READ_RESPONSE_ONLY);
		qp->s_rdma_len -= len;
		bth0 = qp->s_ack_state << 24;
		ohdr->u.aeth = ipath_compute_aeth(qp);
		hwords++;
		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_rdma_sge;
		len = qp->s_rdma_len;
		if (len > pmtu)
			len = pmtu;
		else {
			ohdr->u.aeth = ipath_compute_aeth(qp);
			hwords++;
			qp->s_ack_state = OP(RDMA_READ_RESPONSE_LAST);
		}
		qp->s_rdma_len -= len;
		bth0 = qp->s_ack_state << 24;
		break;

	case OP(RDMA_READ_RESPONSE_LAST):
	case OP(RDMA_READ_RESPONSE_ONLY):
		/*
		 * We have to prevent new requests from changing
		 * the r_sge state while a ipath_verbs_send()
		 * is in progress.
		 */
		qp->s_ack_state = OP(ACKNOWLEDGE);
		bth0 = 0;
		goto bail;

	case OP(COMPARE_SWAP):
	case OP(FETCH_ADD):
		qp->s_cur_sge = NULL;
		len = 0;
		/*
		 * Set the s_ack_state so the receive interrupt handler
		 * won't try to send an ACK (out of order) until this one
		 * is actually sent.
		 */
		qp->s_ack_state = OP(RDMA_READ_RESPONSE_LAST);
		bth0 = OP(ATOMIC_ACKNOWLEDGE) << 24;
		ohdr->u.at.aeth = ipath_compute_aeth(qp);
		ohdr->u.at.atomic_ack_eth = cpu_to_be64(qp->r_atomic_data);
		hwords += sizeof(ohdr->u.at) / 4;
		break;

	default:
		/* Send a regular ACK. */
		qp->s_cur_sge = NULL;
		len = 0;
		/*
		 * Set the s_ack_state so the receive interrupt handler
		 * won't try to send an ACK (out of order) until this one
		 * is actually sent.
		 */
		qp->s_ack_state = OP(RDMA_READ_RESPONSE_LAST);
		bth0 = OP(ACKNOWLEDGE) << 24;
		if (qp->s_nak_state)
			ohdr->u.aeth = cpu_to_be32((qp->r_msn & IPATH_MSN_MASK) |
						    (qp->s_nak_state <<
						     IPATH_AETH_CREDIT_SHIFT));
		else
			ohdr->u.aeth = ipath_compute_aeth(qp);
		hwords++;
	}
	qp->s_hdrwords = hwords;
	qp->s_cur_size = len;

bail:
	return bth0;
}

/**
 * ipath_make_rc_req - construct a request packet (SEND, RDMA r/w, ATOMIC)
 * @qp: a pointer to the QP
 * @ohdr: a pointer to the IB header being constructed
 * @pmtu: the path MTU
 * @bth0p: pointer to the BTH opcode word
 * @bth2p: pointer to the BTH PSN word
 *
 * Return 1 if constructed; otherwise, return 0.
 * Note the QP s_lock must be held and interrupts disabled.
 */
int ipath_make_rc_req(struct ipath_qp *qp,
		      struct ipath_other_headers *ohdr,
		      u32 pmtu, u32 *bth0p, u32 *bth2p)
{
	struct ipath_ibdev *dev = to_idev(qp->ibqp.device);
	struct ipath_sge_state *ss;
	struct ipath_swqe *wqe;
	u32 hwords;
	u32 len;
	u32 bth0;
	u32 bth2;
	char newreq;

	if (!(ib_ipath_state_ops[qp->state] & IPATH_PROCESS_SEND_OK) ||
	    qp->s_rnr_timeout)
		goto done;

	/* header size in 32-bit words LRH+BTH = (8+12)/4. */
	hwords = 5;
	bth0 = 0;

	/* Send a request. */
	wqe = get_swqe_ptr(qp, qp->s_cur);
	switch (qp->s_state) {
	default:
		/*
		 * 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)
				goto done;
			qp->s_psn = wqe->psn = qp->s_next_psn;
			newreq = 1;
		}
		/*
		 * Note that we have to be careful not to modify the
		 * original work request since we may need to resend
		 * it.
		 */
		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_len = len = wqe->length;
		ss = &qp->s_sge;
		bth2 = 0;
		switch (wqe->wr.opcode) {
		case IB_WR_SEND:
		case IB_WR_SEND_WITH_IMM:
			/* If no credit, return. */
			if (qp->s_lsn != (u32) -1 &&
			    ipath_cmp24(wqe->ssn, qp->s_lsn + 1) > 0)
				goto done;
			wqe->lpsn = wqe->psn;
			if (len > pmtu) {
				wqe->lpsn += (len - 1) / pmtu;
				qp->s_state = OP(SEND_FIRST);
				len = pmtu;
				break;
			}
			if (wqe->wr.opcode == IB_WR_SEND)
				qp->s_state = OP(SEND_ONLY);
			else {
				qp->s_state = OP(SEND_ONLY_WITH_IMMEDIATE);
				/* Immediate data comes after the BTH */
				ohdr->u.imm_data = wqe->wr.imm_data;
				hwords += 1;
			}
			if (wqe->wr.send_flags & IB_SEND_SOLICITED)
				bth0 |= 1 << 23;
			bth2 = 1 << 31;	/* Request ACK. */
			if (++qp->s_cur == qp->s_size)
				qp->s_cur = 0;
			break;

		case IB_WR_RDMA_WRITE:
			if (newreq && qp->s_lsn != (u32) -1)
				qp->s_lsn++;
			/* FALLTHROUGH */
		case IB_WR_RDMA_WRITE_WITH_IMM:
			/* If no credit, return. */
			if (qp->s_lsn != (u32) -1 &&
			    ipath_cmp24(wqe->ssn, qp->s_lsn + 1) > 0)
				goto done;
			ohdr->u.rc.reth.vaddr =
				cpu_to_be64(wqe->wr.wr.rdma.remote_addr);
			ohdr->u.rc.reth.rkey =
				cpu_to_be32(wqe->wr.wr.rdma.rkey);
			ohdr->u.rc.reth.length = cpu_to_be32(len);
			hwords += sizeof(struct ib_reth) / 4;
			wqe->lpsn = wqe->psn;
			if (len > pmtu) {
				wqe->lpsn += (len - 1) / 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.imm_data;
				hwords += 1;
				if (wqe->wr.send_flags & IB_SEND_SOLICITED)
					bth0 |= 1 << 23;
			}
			bth2 = 1 << 31;	/* Request ACK. */
			if (++qp->s_cur == qp->s_size)
				qp->s_cur = 0;
			break;

		case IB_WR_RDMA_READ:
			ohdr->u.rc.reth.vaddr =
				cpu_to_be64(wqe->wr.wr.rdma.remote_addr);
			ohdr->u.rc.reth.rkey =
				cpu_to_be32(wqe->wr.wr.rdma.rkey);
			ohdr->u.rc.reth.length = cpu_to_be32(len);
			qp->s_state = OP(RDMA_READ_REQUEST);
			hwords += sizeof(ohdr->u.rc.reth) / 4;
			if (newreq) {
				if (qp->s_lsn != (u32) -1)
					qp->s_lsn++;
				/*
				 * Adjust s_next_psn to count the
				 * expected number of responses.
				 */
				if (len > pmtu)
					qp->s_next_psn += (len - 1) / pmtu;
				wqe->lpsn = qp->s_next_psn++;
			}
			ss = NULL;
			len = 0;
			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:
			if (wqe->wr.opcode == IB_WR_ATOMIC_CMP_AND_SWP)
				qp->s_state = OP(COMPARE_SWAP);
			else
				qp->s_state = OP(FETCH_ADD);
			ohdr->u.atomic_eth.vaddr = cpu_to_be64(
				wqe->wr.wr.atomic.remote_addr);
			ohdr->u.atomic_eth.rkey = cpu_to_be32(
				wqe->wr.wr.atomic.rkey);
			ohdr->u.atomic_eth.swap_data = cpu_to_be64(
				wqe->wr.wr.atomic.swap);
			ohdr->u.atomic_eth.compare_data = cpu_to_be64(
				wqe->wr.wr.atomic.compare_add);
			hwords += sizeof(struct ib_atomic_eth) / 4;
			if (newreq) {
				if (qp->s_lsn != (u32) -1)
					qp->s_lsn++;
				wqe->lpsn = wqe->psn;
			}
			if (++qp->s_cur == qp->s_size)
				qp->s_cur = 0;
			ss = NULL;
			len = 0;
			break;

		default:
			goto done;
		}
		if (newreq) {
			qp->s_tail++;
			if (qp->s_tail >= qp->s_size)
				qp->s_tail = 0;
		}
		bth2 |= qp->s_psn++ & IPATH_PSN_MASK;
		if ((int)(qp->s_psn - qp->s_next_psn) > 0)
			qp->s_next_psn = qp->s_psn;
		/*
		 * Put the QP on the pending list so lost ACKs will cause
		 * a retry.  More than one request can be pending so the
		 * QP may already be on the dev->pending list.
		 */
		spin_lock(&dev->pending_lock);
		if (list_empty(&qp->timerwait))
			list_add_tail(&qp->timerwait,
				      &dev->pending[dev->pending_index]);
		spin_unlock(&dev->pending_lock);
		break;

	case OP(RDMA_READ_RESPONSE_FIRST):
		/*
		 * This case can only happen if a send is restarted.
		 * See ipath_restart_rc().
		 */
		ipath_init_restart(qp, wqe);
		/* FALLTHROUGH */
	case OP(SEND_FIRST):
		qp->s_state = OP(SEND_MIDDLE);
		/* FALLTHROUGH */
	case OP(SEND_MIDDLE):
		bth2 = qp->s_psn++ & IPATH_PSN_MASK;
		if ((int)(qp->s_psn - qp->s_next_psn) > 0)
			qp->s_next_psn = qp->s_psn;
		ss = &qp->s_sge;
		len = qp->s_len;
		if (len > pmtu) {
			/*
			 * Request an ACK every 1/2 MB to avoid retransmit
			 * timeouts.
			 */
			if (((wqe->length - len) % (512 * 1024)) == 0)
				bth2 |= 1 << 31;
			len = pmtu;
			break;
		}
		if (wqe->wr.opcode == IB_WR_SEND)
			qp->s_state = OP(SEND_LAST);
		else {
			qp->s_state = OP(SEND_LAST_WITH_IMMEDIATE);
			/* Immediate data comes after the BTH */
			ohdr->u.imm_data = wqe->wr.imm_data;
			hwords += 1;
		}
		if (wqe->wr.send_flags & IB_SEND_SOLICITED)
			bth0 |= 1 << 23;
		bth2 |= 1 << 31;	/* Request ACK. */
		qp->s_cur++;
		if (qp->s_cur >= qp->s_size)
			qp->s_cur = 0;
		break;

	case OP(RDMA_READ_RESPONSE_LAST):
		/*
		 * This case can only happen if a RDMA write is restarted.
		 * See ipath_restart_rc().
		 */
		ipath_init_restart(qp, wqe);
		/* FALLTHROUGH */
	case OP(RDMA_WRITE_FIRST):
		qp->s_state = OP(RDMA_WRITE_MIDDLE);
		/* FALLTHROUGH */
	case OP(RDMA_WRITE_MIDDLE):
		bth2 = qp->s_psn++ & IPATH_PSN_MASK;
		if ((int)(qp->s_psn - qp->s_next_psn) > 0)
			qp->s_next_psn = qp->s_psn;
		ss = &qp->s_sge;
		len = qp->s_len;
		if (len > pmtu) {
			/*
			 * Request an ACK every 1/2 MB to avoid retransmit
			 * timeouts.
			 */
			if (((wqe->length - len) % (512 * 1024)) == 0)
				bth2 |= 1 << 31;
			len = pmtu;
			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.imm_data;
			hwords += 1;
			if (wqe->wr.send_flags & IB_SEND_SOLICITED)
				bth0 |= 1 << 23;
		}
		bth2 |= 1 << 31;	/* Request ACK. */
		qp->s_cur++;
		if (qp->s_cur >= qp->s_size)
			qp->s_cur = 0;
		break;

	case OP(RDMA_READ_RESPONSE_MIDDLE):
		/*
		 * This case can only happen if a RDMA read is restarted.
		 * See ipath_restart_rc().
		 */
		ipath_init_restart(qp, wqe);
		len = ((qp->s_psn - wqe->psn) & IPATH_PSN_MASK) * pmtu;
		ohdr->u.rc.reth.vaddr =
			cpu_to_be64(wqe->wr.wr.rdma.remote_addr + len);
		ohdr->u.rc.reth.rkey =
			cpu_to_be32(wqe->wr.wr.rdma.rkey);
		ohdr->u.rc.reth.length = cpu_to_be32(qp->s_len);
		qp->s_state = OP(RDMA_READ_REQUEST);
		hwords += sizeof(ohdr->u.rc.reth) / 4;
		bth2 = qp->s_psn++ & IPATH_PSN_MASK;
		if ((int)(qp->s_psn - qp->s_next_psn) > 0)
			qp->s_next_psn = qp->s_psn;
		ss = NULL;
		len = 0;
		qp->s_cur++;
		if (qp->s_cur == qp->s_size)
			qp->s_cur = 0;
		break;

	case OP(RDMA_READ_REQUEST):
	case OP(COMPARE_SWAP):
	case OP(FETCH_ADD):
		/*
		 * We shouldn't start anything new until this request is
		 * finished.  The ACK will handle rescheduling us.  XXX The
		 * number of outstanding ones is negotiated at connection
		 * setup time (see pg. 258,289)?  XXX Also, if we support
		 * multiple outstanding requests, we need to check the WQE
		 * IB_SEND_FENCE flag and not send a new request if a RDMA
		 * read or atomic is pending.
		 */
		goto done;
	}
	qp->s_len -= len;
	qp->s_hdrwords = hwords;
	qp->s_cur_sge = ss;
	qp->s_cur_size = len;
	*bth0p = bth0 | (qp->s_state << 24);
	*bth2p = bth2;
	return 1;

done:
	return 0;
}

/**
 * send_rc_ack - Construct an ACK packet and send it
 * @qp: a pointer to the QP
 *
 * This is called from ipath_rc_rcv() and only uses the receive
 * side QP state.
 * Note that RDMA reads are handled in the send side QP state and tasklet.
 */
static void send_rc_ack(struct ipath_qp *qp)
{
	struct ipath_ibdev *dev = to_idev(qp->ibqp.device);
	u16 lrh0;
	u32 bth0;
	u32 hwords;
	struct ipath_ib_header hdr;
	struct ipath_other_headers *ohdr;

	/* Construct the header. */
	ohdr = &hdr.u.oth;
	lrh0 = IPATH_LRH_BTH;
	/* 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 += ipath_make_grh(dev, &hdr.u.l.grh,
					 &qp->remote_ah_attr.grh,
					 hwords, 0);
		ohdr = &hdr.u.l.oth;
		lrh0 = IPATH_LRH_GRH;
	}
	/* read pkey_index w/o lock (its atomic) */
	bth0 = ipath_layer_get_pkey(dev->dd, qp->s_pkey_index);
	if (qp->r_nak_state)
		ohdr->u.aeth = cpu_to_be32((qp->r_msn & IPATH_MSN_MASK) |
					    (qp->r_nak_state <<
					     IPATH_AETH_CREDIT_SHIFT));
	else
		ohdr->u.aeth = ipath_compute_aeth(qp);
	if (qp->r_ack_state >= OP(COMPARE_SWAP)) {
		bth0 |= OP(ATOMIC_ACKNOWLEDGE) << 24;
		ohdr->u.at.atomic_ack_eth = cpu_to_be64(qp->r_atomic_data);
		hwords += sizeof(ohdr->u.at.atomic_ack_eth) / 4;
	} else
		bth0 |= OP(ACKNOWLEDGE) << 24;
	lrh0 |= qp->remote_ah_attr.sl << 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(ipath_layer_get_lid(dev->dd));
	ohdr->bth[0] = cpu_to_be32(bth0);
	ohdr->bth[1] = cpu_to_be32(qp->remote_qpn);
	ohdr->bth[2] = cpu_to_be32(qp->r_ack_psn & IPATH_PSN_MASK);

	/*
	 * If we can send the ACK, clear the ACK state.
	 */
	if (ipath_verbs_send(dev->dd, hwords, (u32 *) &hdr, 0, NULL) == 0) {
		qp->r_ack_state = OP(ACKNOWLEDGE);
		dev->n_unicast_xmit++;
	} else {
		/*
		 * We are out of PIO buffers at the moment.
		 * Pass responsibility for sending the ACK to the
		 * send tasklet so that when a PIO buffer becomes
		 * available, the ACK is sent ahead of other outgoing
		 * packets.
		 */
		dev->n_rc_qacks++;
		spin_lock_irq(&qp->s_lock);
		/* Don't coalesce if a RDMA read or atomic is pending. */
		if (qp->s_ack_state == OP(ACKNOWLEDGE) ||
		    qp->s_ack_state < OP(RDMA_READ_REQUEST)) {
			qp->s_ack_state = qp->r_ack_state;
			qp->s_nak_state = qp->r_nak_state;
			qp->s_ack_psn = qp->r_ack_psn;
			qp->r_ack_state = OP(ACKNOWLEDGE);
		}
		spin_unlock_irq(&qp->s_lock);

		/* Call ipath_do_rc_send() in another thread. */
		tasklet_hi_schedule(&qp->s_task);
	}
}

/**
 * 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 ipath_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 ipath_qp *qp, u32 psn)
{
	u32 n = qp->s_last;
	struct ipath_swqe *wqe = 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 (ipath_cmp24(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 = get_swqe_ptr(qp, n);
		diff = ipath_cmp24(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 ipath_do_rc_send().
	 */
	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;
}

/**
 * ipath_restart_rc - back up requester to resend the last un-ACKed request
 * @qp: the QP to restart
 * @psn: packet sequence number for the request
 * @wc: the work completion request
 *
 * The QP s_lock should be held and interrupts disabled.
 */
void ipath_restart_rc(struct ipath_qp *qp, u32 psn, struct ib_wc *wc)
{
	struct ipath_swqe *wqe = get_swqe_ptr(qp, qp->s_last);
	struct ipath_ibdev *dev;

	/*
	 * If there are no requests pending, we are done.
	 */
	if (ipath_cmp24(psn, qp->s_next_psn) >= 0 ||
	    qp->s_last == qp->s_tail)
		goto done;

	if (qp->s_retry == 0) {
		wc->wr_id = wqe->wr.wr_id;
		wc->status = IB_WC_RETRY_EXC_ERR;
		wc->opcode = ib_ipath_wc_opcode[wqe->wr.opcode];
		wc->vendor_err = 0;
		wc->byte_len = 0;
		wc->qp_num = qp->ibqp.qp_num;
		wc->src_qp = qp->remote_qpn;
		wc->pkey_index = 0;
		wc->slid = qp->remote_ah_attr.dlid;
		wc->sl = qp->remote_ah_attr.sl;
		wc->dlid_path_bits = 0;
		wc->port_num = 0;
		ipath_sqerror_qp(qp, wc);
		goto bail;
	}
	qp->s_retry--;

	/*
	 * Remove the QP from the timeout queue.
	 * Note: it may already have been removed by ipath_ib_timer().
	 */
	dev = to_idev(qp->ibqp.device);
	spin_lock(&dev->pending_lock);
	if (!list_empty(&qp->timerwait))
		list_del_init(&qp->timerwait);
	spin_unlock(&dev->pending_lock);

	if (wqe->wr.opcode == IB_WR_RDMA_READ)
		dev->n_rc_resends++;
	else
		dev->n_rc_resends += (int)qp->s_psn - (int)psn;

	reset_psn(qp, psn);

done:
	tasklet_hi_schedule(&qp->s_task);

bail:
	return;
}

/**
 * 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 ipath_rc_rcv_resp() to process an incoming RC ACK
 * for the given QP.
 * Called at interrupt level with the QP s_lock held and interrupts disabled.
 * Returns 1 if OK, 0 if current operation should be aborted (NAK).
 */
static int do_rc_ack(struct ipath_qp *qp, u32 aeth, u32 psn, int opcode)
{
	struct ipath_ibdev *dev = to_idev(qp->ibqp.device);
	struct ib_wc wc;
	struct ipath_swqe *wqe;
	int ret = 0;

	/*
	 * Remove the QP from the timeout queue (or RNR timeout queue).
	 * If ipath_ib_timer() has already removed it,
	 * it's OK since we hold the QP s_lock and ipath_restart_rc()
	 * just won't find anything to restart if we ACK everything.
	 */
	spin_lock(&dev->pending_lock);
	if (!list_empty(&qp->timerwait))
		list_del_init(&qp->timerwait);
	spin_unlock(&dev->pending_lock);

	/*
	 * 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).
	 */
	wqe = get_swqe_ptr(qp, qp->s_last);

	/* Nothing is pending to ACK/NAK. */
	if (qp->s_last == qp->s_tail)
		goto bail;

	/*
	 * The MSN might be for a later WQE than the PSN indicates so
	 * only complete WQEs that the PSN finishes.
	 */
	while (ipath_cmp24(psn, wqe->lpsn) >= 0) {
		/* If we are ACKing a WQE, the MSN should be >= the SSN. */
		if (ipath_cmp24(aeth, wqe->ssn) < 0)
			break;
		/*
		 * 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)) ||
		    ((wqe->wr.opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
		      wqe->wr.opcode == IB_WR_ATOMIC_FETCH_AND_ADD) &&
		     (opcode != OP(ATOMIC_ACKNOWLEDGE) ||
		      ipath_cmp24(wqe->psn, psn) != 0))) {
			/*
			 * The last valid PSN seen is the previous
			 * request's.
			 */
			qp->s_last_psn = wqe->psn - 1;
			/* Retry this request. */
			ipath_restart_rc(qp, wqe->psn, &wc);
			/*
			 * No need to process the ACK/NAK since we are
			 * restarting an earlier request.
			 */
			goto bail;
		}
		/* Post a send completion queue entry if requested. */
		if (!test_bit(IPATH_S_SIGNAL_REQ_WR, &qp->s_flags) ||
		    (wqe->wr.send_flags & IB_SEND_SIGNALED)) {
			wc.wr_id = wqe->wr.wr_id;
			wc.status = IB_WC_SUCCESS;
			wc.opcode = ib_ipath_wc_opcode[wqe->wr.opcode];
			wc.vendor_err = 0;
			wc.byte_len = wqe->length;
			wc.qp_num = qp->ibqp.qp_num;
			wc.src_qp = qp->remote_qpn;
			wc.pkey_index = 0;
			wc.slid = qp->remote_ah_attr.dlid;
			wc.sl = qp->remote_ah_attr.sl;
			wc.dlid_path_bits = 0;
			wc.port_num = 0;
			ipath_cq_enter(to_icq(qp->ibqp.send_cq), &wc, 0);
		}
		qp->s_retry = qp->s_retry_cnt;
		/*
		 * If we are completing a request which is in the process of
		 * being resent, we can stop resending it since we know the
		 * responder has already seen it.
		 */
		if (qp->s_last == qp->s_cur) {
			if (++qp->s_cur >= qp->s_size)
				qp->s_cur = 0;
			wqe = get_swqe_ptr(qp, qp->s_cur);
			qp->s_state = OP(SEND_LAST);
			qp->s_psn = wqe->psn;
		}
		if (++qp->s_last >= qp->s_size)
			qp->s_last = 0;
		wqe = get_swqe_ptr(qp, qp->s_last);
		if (qp->s_last == qp->s_tail)
			break;
	}

	switch (aeth >> 29) {
	case 0:		/* ACK */
		dev->n_rc_acks++;
		/* If this is a partial ACK, reset the retransmit timer. */
		if (qp->s_last != qp->s_tail) {
			spin_lock(&dev->pending_lock);
			list_add_tail(&qp->timerwait,
				      &dev->pending[dev->pending_index]);
			spin_unlock(&dev->pending_lock);
		}
		ipath_get_credit(qp, aeth);
		qp->s_rnr_retry = qp->s_rnr_retry_cnt;
		qp->s_retry = qp->s_retry_cnt;
		qp->s_last_psn = psn;
		ret = 1;
		goto bail;

	case 1:		/* RNR NAK */
		dev->n_rnr_naks++;
		if (qp->s_rnr_retry == 0) {
			if (qp->s_last == qp->s_tail)
				goto bail;

			wc.status = IB_WC_RNR_RETRY_EXC_ERR;
			goto class_b;
		}
		if (qp->s_rnr_retry_cnt < 7)
			qp->s_rnr_retry--;
		if (qp->s_last == qp->s_tail)
			goto bail;

		/* The last valid PSN is the previous PSN. */
		qp->s_last_psn = psn - 1;

		dev->n_rc_resends += (int)qp->s_psn - (int)psn;

		reset_psn(qp, psn);

		qp->s_rnr_timeout =
			ib_ipath_rnr_table[(aeth >> IPATH_AETH_CREDIT_SHIFT) &
					   IPATH_AETH_CREDIT_MASK];
		ipath_insert_rnr_queue(qp);
		goto bail;

	case 3:		/* NAK */
		/* The last valid PSN seen is the previous request's. */
		if (qp->s_last != qp->s_tail)
			qp->s_last_psn = wqe->psn - 1;
		switch ((aeth >> IPATH_AETH_CREDIT_SHIFT) &
			IPATH_AETH_CREDIT_MASK) {
		case 0:	/* PSN sequence error */
			dev->n_seq_naks++;
			/*
			 * Back up to the responder's expected PSN.  XXX
			 * Note that we might get a NAK in the middle of an
			 * RDMA READ response which terminates the RDMA
			 * READ.
			 */
			if (qp->s_last == qp->s_tail)
				break;

			if (ipath_cmp24(psn, wqe->psn) < 0)
				break;

			/* Retry the request. */
			ipath_restart_rc(qp, psn, &wc);
			break;

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

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

		case 3:	/* Remote Operation Error */
			wc.status = IB_WC_REM_OP_ERR;
			dev->n_other_naks++;
		class_b:
			wc.wr_id = wqe->wr.wr_id;
			wc.opcode = ib_ipath_wc_opcode[wqe->wr.opcode];
			wc.vendor_err = 0;
			wc.byte_len = 0;
			wc.qp_num = qp->ibqp.qp_num;
			wc.src_qp = qp->remote_qpn;
			wc.pkey_index = 0;
			wc.slid = qp->remote_ah_attr.dlid;
			wc.sl = qp->remote_ah_attr.sl;
			wc.dlid_path_bits = 0;
			wc.port_num = 0;
			ipath_sqerror_qp(qp, &wc);
			break;

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

	default:		/* 2: reserved */
	reserved:
		/* Ignore reserved NAK codes. */
		goto bail;
	}

bail:
	return ret;
}

/**
 * ipath_rc_rcv_resp - process an incoming RC response packet
 * @dev: the device 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
 * @header_in_data: true if part of the header data is in the data buffer
 *
 * This is called from ipath_rc_rcv() to process an incoming RC response
 * packet for the given QP.
 * Called at interrupt level.
 */
static inline void ipath_rc_rcv_resp(struct ipath_ibdev *dev,
				     struct ipath_other_headers *ohdr,
				     void *data, u32 tlen,
				     struct ipath_qp *qp,
				     u32 opcode,
				     u32 psn, u32 hdrsize, u32 pmtu,
				     int header_in_data)
{
	unsigned long flags;
	struct ib_wc wc;
	int diff;
	u32 pad;
	u32 aeth;

	spin_lock_irqsave(&qp->s_lock, flags);

	/* Ignore invalid responses. */
	if (ipath_cmp24(psn, qp->s_next_psn) >= 0)
		goto ack_done;

	/* Ignore duplicate responses. */
	diff = ipath_cmp24(psn, qp->s_last_psn);
	if (unlikely(diff <= 0)) {
		/* Update credits for "ghost" ACKs */
		if (diff == 0 && opcode == OP(ACKNOWLEDGE)) {
			if (!header_in_data)
				aeth = be32_to_cpu(ohdr->u.aeth);
			else {
				aeth = be32_to_cpu(((__be32 *) data)[0]);
				data += sizeof(__be32);
			}
			if ((aeth >> 29) == 0)
				ipath_get_credit(qp, aeth);
		}
		goto ack_done;
	}

	switch (opcode) {
	case OP(ACKNOWLEDGE):
	case OP(ATOMIC_ACKNOWLEDGE):
	case OP(RDMA_READ_RESPONSE_FIRST):
		if (!header_in_data)
			aeth = be32_to_cpu(ohdr->u.aeth);
		else {
			aeth = be32_to_cpu(((__be32 *) data)[0]);
			data += sizeof(__be32);
		}
		if (opcode == OP(ATOMIC_ACKNOWLEDGE))
			*(u64 *) qp->s_sge.sge.vaddr = *(u64 *) data;
		if (!do_rc_ack(qp, aeth, psn, opcode) ||
		    opcode != OP(RDMA_READ_RESPONSE_FIRST))
			goto ack_done;
		hdrsize += 4;
		/*
		 * do_rc_ack() has already checked the PSN so skip
		 * the sequence check.
		 */
		goto rdma_read;

	case OP(RDMA_READ_RESPONSE_MIDDLE):
		/* no AETH, no ACK */
		if (unlikely(ipath_cmp24(psn, qp->s_last_psn + 1))) {
			dev->n_rdma_seq++;
			ipath_restart_rc(qp, qp->s_last_psn + 1, &wc);
			goto ack_done;
		}
	rdma_read:
		if (unlikely(qp->s_state != OP(RDMA_READ_REQUEST)))
			goto ack_done;
		if (unlikely(tlen != (hdrsize + pmtu + 4)))
			goto ack_done;
		if (unlikely(pmtu >= qp->s_len))
			goto ack_done;
		/* We got a response so update the timeout. */
		if (unlikely(qp->s_last == qp->s_tail ||
			     get_swqe_ptr(qp, qp->s_last)->wr.opcode !=
			     IB_WR_RDMA_READ))
			goto ack_done;
		spin_lock(&dev->pending_lock);
		if (qp->s_rnr_timeout == 0 && !list_empty(&qp->timerwait))
			list_move_tail(&qp->timerwait,
				       &dev->pending[dev->pending_index]);
		spin_unlock(&dev->pending_lock);
		/*
		 * Update the RDMA receive state but do the copy w/o
		 * holding the locks and blocking interrupts.
		 * XXX Yet another place that affects relaxed RDMA order
		 * since we don't want s_sge modified.
		 */
		qp->s_len -= pmtu;
		qp->s_last_psn = psn;
		spin_unlock_irqrestore(&qp->s_lock, flags);
		ipath_copy_sge(&qp->s_sge, data, pmtu);
		goto bail;

	case OP(RDMA_READ_RESPONSE_LAST):
		/* ACKs READ req. */
		if (unlikely(ipath_cmp24(psn, qp->s_last_psn + 1))) {
			dev->n_rdma_seq++;
			ipath_restart_rc(qp, qp->s_last_psn + 1, &wc);
			goto ack_done;
		}
		/* FALLTHROUGH */
	case OP(RDMA_READ_RESPONSE_ONLY):
		if (unlikely(qp->s_state != OP(RDMA_READ_REQUEST)))
			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 >= 1 && <= pmtu.
		 * Remember to account for the AETH header (4) and
		 * ICRC (4).
		 */
		if (unlikely(tlen <= (hdrsize + pad + 8))) {
			/* XXX Need to generate an error CQ entry. */
			goto ack_done;
		}
		tlen -= hdrsize + pad + 8;
		if (unlikely(tlen != qp->s_len)) {
			/* XXX Need to generate an error CQ entry. */
			goto ack_done;
		}
		if (!header_in_data)
			aeth = be32_to_cpu(ohdr->u.aeth);
		else {
			aeth = be32_to_cpu(((__be32 *) data)[0]);
			data += sizeof(__be32);
		}
		ipath_copy_sge(&qp->s_sge, data, tlen);
		if (do_rc_ack(qp, aeth, psn, OP(RDMA_READ_RESPONSE_LAST))) {
			/*
			 * Change the state so we contimue
			 * processing new requests and wake up the
			 * tasklet if there are posted sends.
			 */
			qp->s_state = OP(SEND_LAST);
			if (qp->s_tail != qp->s_head)
				tasklet_hi_schedule(&qp->s_task);
		}
		goto ack_done;
	}

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

/**
 * ipath_rc_rcv_error - process an incoming duplicate or error RC packet
 * @dev: the device this packet came in on
 * @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
 * @header_in_data: true if part of the header data is in the data buffer
 *
 * This is called from ipath_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 and the s_lock unlocked.
 */
static inline int ipath_rc_rcv_error(struct ipath_ibdev *dev,
				     struct ipath_other_headers *ohdr,
				     void *data,
				     struct ipath_qp *qp,
				     u32 opcode,
				     u32 psn,
				     int diff,
				     int header_in_data)
{
	struct ib_reth *reth;

	if (diff > 0) {
		/*
		 * Packet sequence error.
		 * A NAK will ACK earlier sends and RDMA writes.
		 * Don't queue the NAK if a RDMA read, atomic, or
		 * NAK is pending though.
		 */
		if (qp->s_ack_state != OP(ACKNOWLEDGE) ||
		    qp->r_nak_state != 0)
			goto done;
		if (qp->r_ack_state < OP(COMPARE_SWAP)) {
			qp->r_ack_state = OP(SEND_ONLY);
			qp->r_nak_state = IB_NAK_PSN_ERROR;
			/* Use the expected PSN. */
			qp->r_ack_psn = qp->r_psn;
		}
		goto send_ack;
	}

	/*
	 * 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.
	 */
	if (opcode == OP(RDMA_READ_REQUEST)) {
		/* RETH comes after BTH */
		if (!header_in_data)
			reth = &ohdr->u.rc.reth;
		else {
			reth = (struct ib_reth *)data;
			data += sizeof(*reth);
		}
		/*
		 * If we receive a duplicate RDMA request, it means the
		 * requester saw a sequence error and needs to restart
		 * from an earlier point.  We can abort the current
		 * RDMA read send in that case.
		 */
		spin_lock_irq(&qp->s_lock);
		if (qp->s_ack_state != OP(ACKNOWLEDGE) &&
		    (qp->s_hdrwords || ipath_cmp24(psn, qp->s_ack_psn) >= 0)) {
			/*
			 * We are already sending earlier requested data.
			 * Don't abort it to send later out of sequence data.
			 */
			spin_unlock_irq(&qp->s_lock);
			goto done;
		}
		qp->s_rdma_len = be32_to_cpu(reth->length);
		if (qp->s_rdma_len != 0) {
			u32 rkey = be32_to_cpu(reth->rkey);
			u64 vaddr = be64_to_cpu(reth->vaddr);
			int ok;

			/*
			 * Address range must be a subset of the original
			 * request and start on pmtu boundaries.
			 */
			ok = ipath_rkey_ok(dev, &qp->s_rdma_sge,
					   qp->s_rdma_len, vaddr, rkey,
					   IB_ACCESS_REMOTE_READ);
			if (unlikely(!ok)) {
				spin_unlock_irq(&qp->s_lock);
				goto done;
			}
		} else {
			qp->s_rdma_sge.sg_list = NULL;
			qp->s_rdma_sge.num_sge = 0;
			qp->s_rdma_sge.sge.mr = NULL;
			qp->s_rdma_sge.sge.vaddr = NULL;
			qp->s_rdma_sge.sge.length = 0;
			qp->s_rdma_sge.sge.sge_length = 0;
		}
		qp->s_ack_state = opcode;
		qp->s_ack_psn = psn;
		spin_unlock_irq(&qp->s_lock);
		tasklet_hi_schedule(&qp->s_task);
		goto send_ack;
	}

	/*
	 * A pending RDMA read will ACK anything before it so
	 * ignore earlier duplicate requests.
	 */
	if (qp->s_ack_state != OP(ACKNOWLEDGE))
		goto done;

	/*
	 * If an ACK is pending, don't replace the pending ACK
	 * with an earlier one since the later one will ACK the earlier.
	 * Also, if we already have a pending atomic, send it.
	 */
	if (qp->r_ack_state != OP(ACKNOWLEDGE) &&
	    (ipath_cmp24(psn, qp->r_ack_psn) <= 0 ||
	     qp->r_ack_state >= OP(COMPARE_SWAP)))
		goto send_ack;
	switch (opcode) {
	case OP(COMPARE_SWAP):
	case OP(FETCH_ADD):
		/*
		 * Check for the PSN of the last atomic operation
		 * performed and resend the result if found.
		 */
		if ((psn & IPATH_PSN_MASK) != qp->r_atomic_psn)
			goto done;
		break;
	}
	qp->r_ack_state = opcode;
	qp->r_nak_state = 0;
	qp->r_ack_psn = psn;
send_ack:
	return 0;

done:
	return 1;
}

/**
 * ipath_rc_rcv - process an incoming RC packet
 * @dev: the device this packet came in on
 * @hdr: the header of this packet
 * @has_grh: true if the header has a GRH
 * @data: the packet data
 * @tlen: the packet length
 * @qp: the QP for this packet
 *
 * This is called from ipath_qp_rcv() to process an incoming RC packet
 * for the given QP.
 * Called at interrupt level.
 */
void ipath_rc_rcv(struct ipath_ibdev *dev, struct ipath_ib_header *hdr,
		  int has_grh, void *data, u32 tlen, struct ipath_qp *qp)
{
	struct ipath_other_headers *ohdr;
	u32 opcode;
	u32 hdrsize;
	u32 psn;
	u32 pad;
	struct ib_wc wc;
	u32 pmtu = ib_mtu_enum_to_int(qp->path_mtu);
	int diff;
	struct ib_reth *reth;
	int header_in_data;

	/* Check for GRH */
	if (!has_grh) {
		ohdr = &hdr->u.oth;
		hdrsize = 8 + 12;	/* LRH + BTH */
		psn = be32_to_cpu(ohdr->bth[2]);
		header_in_data = 0;
	} else {
		ohdr = &hdr->u.l.oth;
		hdrsize = 8 + 40 + 12;	/* LRH + GRH + BTH */
		/*
		 * The header with GRH is 60 bytes and the core driver sets
		 * the eager header buffer size to 56 bytes so the last 4
		 * bytes of the BTH header (PSN) is in the data buffer.
		 */
		header_in_data =
			ipath_layer_get_rcvhdrentsize(dev->dd) == 16;
		if (header_in_data) {
			psn = be32_to_cpu(((__be32 *) data)[0]);
			data += sizeof(__be32);
		} else
			psn = be32_to_cpu(ohdr->bth[2]);
	}

	/*
	 * 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.
	 */
	opcode = be32_to_cpu(ohdr->bth[0]) >> 24;
	if (opcode >= OP(RDMA_READ_RESPONSE_FIRST) &&
	    opcode <= OP(ATOMIC_ACKNOWLEDGE)) {
		ipath_rc_rcv_resp(dev, ohdr, data, tlen, qp, opcode, psn,
				  hdrsize, pmtu, header_in_data);
		goto done;
	}

	/* Compute 24 bits worth of difference. */
	diff = ipath_cmp24(psn, qp->r_psn);
	if (unlikely(diff)) {
		if (ipath_rc_rcv_error(dev, ohdr, data, qp, opcode,
				       psn, diff, header_in_data))
			goto done;
		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))
			break;
	nack_inv:
		/*
		 * A NAK will ACK earlier sends and RDMA writes.
		 * Don't queue the NAK if a RDMA read, atomic, or NAK
		 * is pending though.
		 */
		if (qp->r_ack_state >= OP(COMPARE_SWAP))
			goto send_ack;
		/* XXX Flush WQEs */
		qp->state = IB_QPS_ERR;
		qp->r_ack_state = OP(SEND_ONLY);
		qp->r_nak_state = IB_NAK_INVALID_REQUEST;
		qp->r_ack_psn = qp->r_psn;
		goto send_ack;

	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(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;
	}

	wc.imm_data = 0;
	wc.wc_flags = 0;

	/* OK, process the packet. */
	switch (opcode) {
	case OP(SEND_FIRST):
		if (!ipath_get_rwqe(qp, 0)) {
		rnr_nak:
			/*
			 * A RNR NAK will ACK earlier sends and RDMA writes.
			 * Don't queue the NAK if a RDMA read or atomic
			 * is pending though.
			 */
			if (qp->r_ack_state >= OP(COMPARE_SWAP))
				goto send_ack;
			qp->r_ack_state = OP(SEND_ONLY);
			qp->r_nak_state = IB_RNR_NAK | qp->r_min_rnr_timer;
			qp->r_ack_psn = qp->r_psn;
			goto send_ack;
		}
		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;
		ipath_copy_sge(&qp->r_sge, data, pmtu);
		break;

	case OP(RDMA_WRITE_LAST_WITH_IMMEDIATE):
		/* consume RWQE */
		if (!ipath_get_rwqe(qp, 1))
			goto rnr_nak;
		goto send_last_imm;

	case OP(SEND_ONLY):
	case OP(SEND_ONLY_WITH_IMMEDIATE):
		if (!ipath_get_rwqe(qp, 0))
			goto rnr_nak;
		qp->r_rcv_len = 0;
		if (opcode == OP(SEND_ONLY))
			goto send_last;
		/* FALLTHROUGH */
	case OP(SEND_LAST_WITH_IMMEDIATE):
	send_last_imm:
		if (header_in_data) {
			wc.imm_data = *(__be32 *) data;
			data += sizeof(__be32);
		} else {
			/* Immediate data comes after BTH */
			wc.imm_data = ohdr->u.imm_data;
		}
		hdrsize += 4;
		wc.wc_flags = IB_WC_WITH_IMM;
		/* FALLTHROUGH */
	case OP(SEND_LAST):
	case OP(RDMA_WRITE_LAST):
	send_last:
		/* Get the number of bytes the message was padded by. */
		pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
		/* Check for invalid length. */
		/* XXX 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;
		ipath_copy_sge(&qp->r_sge, data, tlen);
		qp->r_msn++;
		if (opcode == OP(RDMA_WRITE_LAST) ||
		    opcode == OP(RDMA_WRITE_ONLY))
			break;
		wc.wr_id = qp->r_wr_id;
		wc.status = IB_WC_SUCCESS;
		wc.opcode = IB_WC_RECV;
		wc.vendor_err = 0;
		wc.qp_num = qp->ibqp.qp_num;
		wc.src_qp = qp->remote_qpn;
		wc.pkey_index = 0;
		wc.slid = qp->remote_ah_attr.dlid;
		wc.sl = qp->remote_ah_attr.sl;
		wc.dlid_path_bits = 0;
		wc.port_num = 0;
		/* Signal completion event if the solicited bit is set. */
		ipath_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
			       (ohdr->bth[0] &
				__constant_cpu_to_be32(1 << 23)) != 0);
		break;

	case OP(RDMA_WRITE_FIRST):
	case OP(RDMA_WRITE_ONLY):
	case OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE):
		/* consume RWQE */
		/* RETH comes after BTH */
		if (!header_in_data)
			reth = &ohdr->u.rc.reth;
		else {
			reth = (struct ib_reth *)data;
			data += sizeof(*reth);
		}
		hdrsize += sizeof(*reth);
		qp->r_len = be32_to_cpu(reth->length);
		qp->r_rcv_len = 0;
		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 = ipath_rkey_ok(dev, &qp->r_sge,
					   qp->r_len, vaddr, rkey,
					   IB_ACCESS_REMOTE_WRITE);
			if (unlikely(!ok))
				goto nack_acc;
		} else {
			qp->r_sge.sg_list = NULL;
			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 (unlikely(!(qp->qp_access_flags &
			       IB_ACCESS_REMOTE_WRITE)))
			goto nack_acc;
		if (opcode == OP(RDMA_WRITE_FIRST))
			goto send_middle;
		else if (opcode == OP(RDMA_WRITE_ONLY))
			goto send_last;
		if (!ipath_get_rwqe(qp, 1))
			goto rnr_nak;
		goto send_last_imm;

	case OP(RDMA_READ_REQUEST):
		/* RETH comes after BTH */
		if (!header_in_data)
			reth = &ohdr->u.rc.reth;
		else {
			reth = (struct ib_reth *)data;
			data += sizeof(*reth);
		}
		if (unlikely(!(qp->qp_access_flags &
			       IB_ACCESS_REMOTE_READ)))
			goto nack_acc;
		spin_lock_irq(&qp->s_lock);
		qp->s_rdma_len = be32_to_cpu(reth->length);
		if (qp->s_rdma_len != 0) {
			u32 rkey = be32_to_cpu(reth->rkey);
			u64 vaddr = be64_to_cpu(reth->vaddr);
			int ok;

			/* Check rkey & NAK */
			ok = ipath_rkey_ok(dev, &qp->s_rdma_sge,
					   qp->s_rdma_len, vaddr, rkey,
					   IB_ACCESS_REMOTE_READ);
			if (unlikely(!ok)) {
				spin_unlock_irq(&qp->s_lock);
				goto nack_acc;
			}
			/*
			 * Update the next expected PSN.  We add 1 later
			 * below, so only add the remainder here.
			 */
			if (qp->s_rdma_len > pmtu)
				qp->r_psn += (qp->s_rdma_len - 1) / pmtu;
		} else {
			qp->s_rdma_sge.sg_list = NULL;
			qp->s_rdma_sge.num_sge = 0;
			qp->s_rdma_sge.sge.mr = NULL;
			qp->s_rdma_sge.sge.vaddr = NULL;
			qp->s_rdma_sge.sge.length = 0;
			qp->s_rdma_sge.sge.sge_length = 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->s_ack_state = opcode;
		qp->s_ack_psn = psn;
		spin_unlock_irq(&qp->s_lock);

		qp->r_psn++;
		qp->r_state = opcode;
		qp->r_nak_state = 0;

		/* Call ipath_do_rc_send() in another thread. */
		tasklet_hi_schedule(&qp->s_task);

		goto done;

	case OP(COMPARE_SWAP):
	case OP(FETCH_ADD): {
		struct ib_atomic_eth *ateth;
		u64 vaddr;
		u64 sdata;
		u32 rkey;

		if (!header_in_data)
			ateth = &ohdr->u.atomic_eth;
		else {
			ateth = (struct ib_atomic_eth *)data;
			data += sizeof(*ateth);
		}
		vaddr = be64_to_cpu(ateth->vaddr);
		if (unlikely(vaddr & (sizeof(u64) - 1)))
			goto nack_inv;
		rkey = be32_to_cpu(ateth->rkey);
		/* Check rkey & NAK */
		if (unlikely(!ipath_rkey_ok(dev, &qp->r_sge,
					    sizeof(u64), vaddr, rkey,
					    IB_ACCESS_REMOTE_ATOMIC)))
			goto nack_acc;
		if (unlikely(!(qp->qp_access_flags &
			       IB_ACCESS_REMOTE_ATOMIC)))
			goto nack_acc;
		/* Perform atomic OP and save result. */
		sdata = be64_to_cpu(ateth->swap_data);
		spin_lock_irq(&dev->pending_lock);
		qp->r_atomic_data = *(u64 *) qp->r_sge.sge.vaddr;
		if (opcode == OP(FETCH_ADD))
			*(u64 *) qp->r_sge.sge.vaddr =
				qp->r_atomic_data + sdata;
		else if (qp->r_atomic_data ==
			 be64_to_cpu(ateth->compare_data))
			*(u64 *) qp->r_sge.sge.vaddr = sdata;
		spin_unlock_irq(&dev->pending_lock);
		qp->r_msn++;
		qp->r_atomic_psn = psn & IPATH_PSN_MASK;
		psn |= 1 << 31;
		break;
	}

	default:
		/* Drop packet for unknown opcodes. */
		goto done;
	}
	qp->r_psn++;
	qp->r_state = opcode;
	qp->r_nak_state = 0;
	/* Send an ACK if requested or required. */
	if (psn & (1 << 31)) {
		/*
		 * Coalesce ACKs unless there is a RDMA READ or
		 * ATOMIC pending.
		 */
		if (qp->r_ack_state < OP(COMPARE_SWAP)) {
			qp->r_ack_state = opcode;
			qp->r_ack_psn = psn;
		}
		goto send_ack;
	}
	goto done;

nack_acc:
	/*
	 * A NAK will ACK earlier sends and RDMA writes.
	 * Don't queue the NAK if a RDMA read, atomic, or NAK
	 * is pending though.
	 */
	if (qp->r_ack_state < OP(COMPARE_SWAP)) {
		/* XXX Flush WQEs */
		qp->state = IB_QPS_ERR;
		qp->r_ack_state = OP(RDMA_WRITE_ONLY);
		qp->r_nak_state = IB_NAK_REMOTE_ACCESS_ERROR;
		qp->r_ack_psn = qp->r_psn;
	}
send_ack:
	/* Send ACK right away unless the send tasklet has a pending ACK. */
	if (qp->s_ack_state == OP(ACKNOWLEDGE))
		send_rc_ack(qp);

done:
	return;
}