summaryrefslogtreecommitdiff
path: root/drivers/infiniband/hw/qib/qib_init.c
blob: b5a78576c48bd79dbe5421882897aa162b9a5b7c (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
/*
 * Copyright (c) 2012, 2013 Intel Corporation.  All rights reserved.
 * Copyright (c) 2006 - 2012 QLogic Corporation. All rights reserved.
 * Copyright (c) 2003, 2004, 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 <linux/pci.h>
#include <linux/netdevice.h>
#include <linux/vmalloc.h>
#include <linux/delay.h>
#include <linux/module.h>
#include <linux/printk.h>
#ifdef CONFIG_INFINIBAND_QIB_DCA
#include <linux/dca.h>
#endif
#include <rdma/rdma_vt.h>

#include "qib.h"
#include "qib_common.h"
#include "qib_mad.h"
#ifdef CONFIG_DEBUG_FS
#include "qib_debugfs.h"
#include "qib_verbs.h"
#endif

#undef pr_fmt
#define pr_fmt(fmt) QIB_DRV_NAME ": " fmt

/*
 * min buffers we want to have per context, after driver
 */
#define QIB_MIN_USER_CTXT_BUFCNT 7

#define QLOGIC_IB_R_SOFTWARE_MASK 0xFF
#define QLOGIC_IB_R_SOFTWARE_SHIFT 24
#define QLOGIC_IB_R_EMULATOR_MASK (1ULL<<62)

/*
 * Number of ctxts we are configured to use (to allow for more pio
 * buffers per ctxt, etc.)  Zero means use chip value.
 */
ushort qib_cfgctxts;
module_param_named(cfgctxts, qib_cfgctxts, ushort, S_IRUGO);
MODULE_PARM_DESC(cfgctxts, "Set max number of contexts to use");

unsigned qib_numa_aware;
module_param_named(numa_aware, qib_numa_aware, uint, S_IRUGO);
MODULE_PARM_DESC(numa_aware,
	"0 -> PSM allocation close to HCA, 1 -> PSM allocation local to process");

/*
 * If set, do not write to any regs if avoidable, hack to allow
 * check for deranged default register values.
 */
ushort qib_mini_init;
module_param_named(mini_init, qib_mini_init, ushort, S_IRUGO);
MODULE_PARM_DESC(mini_init, "If set, do minimal diag init");

unsigned qib_n_krcv_queues;
module_param_named(krcvqs, qib_n_krcv_queues, uint, S_IRUGO);
MODULE_PARM_DESC(krcvqs, "number of kernel receive queues per IB port");

unsigned qib_cc_table_size;
module_param_named(cc_table_size, qib_cc_table_size, uint, S_IRUGO);
MODULE_PARM_DESC(cc_table_size, "Congestion control table entries 0 (CCA disabled - default), min = 128, max = 1984");

static void verify_interrupt(struct timer_list *);

DEFINE_XARRAY_FLAGS(qib_dev_table, XA_FLAGS_ALLOC | XA_FLAGS_LOCK_IRQ);
u32 qib_cpulist_count;
unsigned long *qib_cpulist;

/* set number of contexts we'll actually use */
void qib_set_ctxtcnt(struct qib_devdata *dd)
{
	if (!qib_cfgctxts) {
		dd->cfgctxts = dd->first_user_ctxt + num_online_cpus();
		if (dd->cfgctxts > dd->ctxtcnt)
			dd->cfgctxts = dd->ctxtcnt;
	} else if (qib_cfgctxts < dd->num_pports)
		dd->cfgctxts = dd->ctxtcnt;
	else if (qib_cfgctxts <= dd->ctxtcnt)
		dd->cfgctxts = qib_cfgctxts;
	else
		dd->cfgctxts = dd->ctxtcnt;
	dd->freectxts = (dd->first_user_ctxt > dd->cfgctxts) ? 0 :
		dd->cfgctxts - dd->first_user_ctxt;
}

/*
 * Common code for creating the receive context array.
 */
int qib_create_ctxts(struct qib_devdata *dd)
{
	unsigned i;
	int local_node_id = pcibus_to_node(dd->pcidev->bus);

	if (local_node_id < 0)
		local_node_id = numa_node_id();
	dd->assigned_node_id = local_node_id;

	/*
	 * Allocate full ctxtcnt array, rather than just cfgctxts, because
	 * cleanup iterates across all possible ctxts.
	 */
	dd->rcd = kcalloc(dd->ctxtcnt, sizeof(*dd->rcd), GFP_KERNEL);
	if (!dd->rcd)
		return -ENOMEM;

	/* create (one or more) kctxt */
	for (i = 0; i < dd->first_user_ctxt; ++i) {
		struct qib_pportdata *ppd;
		struct qib_ctxtdata *rcd;

		if (dd->skip_kctxt_mask & (1 << i))
			continue;

		ppd = dd->pport + (i % dd->num_pports);

		rcd = qib_create_ctxtdata(ppd, i, dd->assigned_node_id);
		if (!rcd) {
			qib_dev_err(dd,
				"Unable to allocate ctxtdata for Kernel ctxt, failing\n");
			kfree(dd->rcd);
			dd->rcd = NULL;
			return -ENOMEM;
		}
		rcd->pkeys[0] = QIB_DEFAULT_P_KEY;
		rcd->seq_cnt = 1;
	}
	return 0;
}

/*
 * Common code for user and kernel context setup.
 */
struct qib_ctxtdata *qib_create_ctxtdata(struct qib_pportdata *ppd, u32 ctxt,
	int node_id)
{
	struct qib_devdata *dd = ppd->dd;
	struct qib_ctxtdata *rcd;

	rcd = kzalloc_node(sizeof(*rcd), GFP_KERNEL, node_id);
	if (rcd) {
		INIT_LIST_HEAD(&rcd->qp_wait_list);
		rcd->node_id = node_id;
		rcd->ppd = ppd;
		rcd->dd = dd;
		rcd->cnt = 1;
		rcd->ctxt = ctxt;
		dd->rcd[ctxt] = rcd;
#ifdef CONFIG_DEBUG_FS
		if (ctxt < dd->first_user_ctxt) { /* N/A for PSM contexts */
			rcd->opstats = kzalloc_node(sizeof(*rcd->opstats),
				GFP_KERNEL, node_id);
			if (!rcd->opstats) {
				kfree(rcd);
				qib_dev_err(dd,
					"Unable to allocate per ctxt stats buffer\n");
				return NULL;
			}
		}
#endif
		dd->f_init_ctxt(rcd);

		/*
		 * To avoid wasting a lot of memory, we allocate 32KB chunks
		 * of physically contiguous memory, advance through it until
		 * used up and then allocate more.  Of course, we need
		 * memory to store those extra pointers, now.  32KB seems to
		 * be the most that is "safe" under memory pressure
		 * (creating large files and then copying them over
		 * NFS while doing lots of MPI jobs).  The OOM killer can
		 * get invoked, even though we say we can sleep and this can
		 * cause significant system problems....
		 */
		rcd->rcvegrbuf_size = 0x8000;
		rcd->rcvegrbufs_perchunk =
			rcd->rcvegrbuf_size / dd->rcvegrbufsize;
		rcd->rcvegrbuf_chunks = (rcd->rcvegrcnt +
			rcd->rcvegrbufs_perchunk - 1) /
			rcd->rcvegrbufs_perchunk;
		rcd->rcvegrbufs_perchunk_shift =
			ilog2(rcd->rcvegrbufs_perchunk);
	}
	return rcd;
}

/*
 * Common code for initializing the physical port structure.
 */
int qib_init_pportdata(struct qib_pportdata *ppd, struct qib_devdata *dd,
			u8 hw_pidx, u8 port)
{
	int size;

	ppd->dd = dd;
	ppd->hw_pidx = hw_pidx;
	ppd->port = port; /* IB port number, not index */

	spin_lock_init(&ppd->sdma_lock);
	spin_lock_init(&ppd->lflags_lock);
	spin_lock_init(&ppd->cc_shadow_lock);
	init_waitqueue_head(&ppd->state_wait);

	timer_setup(&ppd->symerr_clear_timer, qib_clear_symerror_on_linkup, 0);

	ppd->qib_wq = NULL;
	ppd->ibport_data.pmastats =
		alloc_percpu(struct qib_pma_counters);
	if (!ppd->ibport_data.pmastats)
		return -ENOMEM;
	ppd->ibport_data.rvp.rc_acks = alloc_percpu(u64);
	ppd->ibport_data.rvp.rc_qacks = alloc_percpu(u64);
	ppd->ibport_data.rvp.rc_delayed_comp = alloc_percpu(u64);
	if (!(ppd->ibport_data.rvp.rc_acks) ||
	    !(ppd->ibport_data.rvp.rc_qacks) ||
	    !(ppd->ibport_data.rvp.rc_delayed_comp))
		return -ENOMEM;

	if (qib_cc_table_size < IB_CCT_MIN_ENTRIES)
		goto bail;

	ppd->cc_supported_table_entries = min(max_t(int, qib_cc_table_size,
		IB_CCT_MIN_ENTRIES), IB_CCT_ENTRIES*IB_CC_TABLE_CAP_DEFAULT);

	ppd->cc_max_table_entries =
		ppd->cc_supported_table_entries/IB_CCT_ENTRIES;

	size = IB_CC_TABLE_CAP_DEFAULT * sizeof(struct ib_cc_table_entry)
		* IB_CCT_ENTRIES;
	ppd->ccti_entries = kzalloc(size, GFP_KERNEL);
	if (!ppd->ccti_entries)
		goto bail;

	size = IB_CC_CCS_ENTRIES * sizeof(struct ib_cc_congestion_entry);
	ppd->congestion_entries = kzalloc(size, GFP_KERNEL);
	if (!ppd->congestion_entries)
		goto bail_1;

	size = sizeof(struct cc_table_shadow);
	ppd->ccti_entries_shadow = kzalloc(size, GFP_KERNEL);
	if (!ppd->ccti_entries_shadow)
		goto bail_2;

	size = sizeof(struct ib_cc_congestion_setting_attr);
	ppd->congestion_entries_shadow = kzalloc(size, GFP_KERNEL);
	if (!ppd->congestion_entries_shadow)
		goto bail_3;

	return 0;

bail_3:
	kfree(ppd->ccti_entries_shadow);
	ppd->ccti_entries_shadow = NULL;
bail_2:
	kfree(ppd->congestion_entries);
	ppd->congestion_entries = NULL;
bail_1:
	kfree(ppd->ccti_entries);
	ppd->ccti_entries = NULL;
bail:
	/* User is intentionally disabling the congestion control agent */
	if (!qib_cc_table_size)
		return 0;

	if (qib_cc_table_size < IB_CCT_MIN_ENTRIES) {
		qib_cc_table_size = 0;
		qib_dev_err(dd,
		 "Congestion Control table size %d less than minimum %d for port %d\n",
		 qib_cc_table_size, IB_CCT_MIN_ENTRIES, port);
	}

	qib_dev_err(dd, "Congestion Control Agent disabled for port %d\n",
		port);
	return 0;
}

static int init_pioavailregs(struct qib_devdata *dd)
{
	int ret, pidx;
	u64 *status_page;

	dd->pioavailregs_dma = dma_alloc_coherent(
		&dd->pcidev->dev, PAGE_SIZE, &dd->pioavailregs_phys,
		GFP_KERNEL);
	if (!dd->pioavailregs_dma) {
		qib_dev_err(dd,
			"failed to allocate PIOavail reg area in memory\n");
		ret = -ENOMEM;
		goto done;
	}

	/*
	 * We really want L2 cache aligned, but for current CPUs of
	 * interest, they are the same.
	 */
	status_page = (u64 *)
		((char *) dd->pioavailregs_dma +
		 ((2 * L1_CACHE_BYTES +
		   dd->pioavregs * sizeof(u64)) & ~L1_CACHE_BYTES));
	/* device status comes first, for backwards compatibility */
	dd->devstatusp = status_page;
	*status_page++ = 0;
	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
		dd->pport[pidx].statusp = status_page;
		*status_page++ = 0;
	}

	/*
	 * Setup buffer to hold freeze and other messages, accessible to
	 * apps, following statusp.  This is per-unit, not per port.
	 */
	dd->freezemsg = (char *) status_page;
	*dd->freezemsg = 0;
	/* length of msg buffer is "whatever is left" */
	ret = (char *) status_page - (char *) dd->pioavailregs_dma;
	dd->freezelen = PAGE_SIZE - ret;

	ret = 0;

done:
	return ret;
}

/**
 * init_shadow_tids - allocate the shadow TID array
 * @dd: the qlogic_ib device
 *
 * allocate the shadow TID array, so we can qib_munlock previous
 * entries.  It may make more sense to move the pageshadow to the
 * ctxt data structure, so we only allocate memory for ctxts actually
 * in use, since we at 8k per ctxt, now.
 * We don't want failures here to prevent use of the driver/chip,
 * so no return value.
 */
static void init_shadow_tids(struct qib_devdata *dd)
{
	struct page **pages;
	dma_addr_t *addrs;

	pages = vzalloc(array_size(sizeof(struct page *),
				   dd->cfgctxts * dd->rcvtidcnt));
	if (!pages)
		goto bail;

	addrs = vzalloc(array_size(sizeof(dma_addr_t),
				   dd->cfgctxts * dd->rcvtidcnt));
	if (!addrs)
		goto bail_free;

	dd->pageshadow = pages;
	dd->physshadow = addrs;
	return;

bail_free:
	vfree(pages);
bail:
	dd->pageshadow = NULL;
}

/*
 * Do initialization for device that is only needed on
 * first detect, not on resets.
 */
static int loadtime_init(struct qib_devdata *dd)
{
	int ret = 0;

	if (((dd->revision >> QLOGIC_IB_R_SOFTWARE_SHIFT) &
	     QLOGIC_IB_R_SOFTWARE_MASK) != QIB_CHIP_SWVERSION) {
		qib_dev_err(dd,
			"Driver only handles version %d, chip swversion is %d (%llx), failing\n",
			QIB_CHIP_SWVERSION,
			(int)(dd->revision >>
				QLOGIC_IB_R_SOFTWARE_SHIFT) &
				QLOGIC_IB_R_SOFTWARE_MASK,
			(unsigned long long) dd->revision);
		ret = -ENOSYS;
		goto done;
	}

	if (dd->revision & QLOGIC_IB_R_EMULATOR_MASK)
		qib_devinfo(dd->pcidev, "%s", dd->boardversion);

	spin_lock_init(&dd->pioavail_lock);
	spin_lock_init(&dd->sendctrl_lock);
	spin_lock_init(&dd->uctxt_lock);
	spin_lock_init(&dd->qib_diag_trans_lock);
	spin_lock_init(&dd->eep_st_lock);
	mutex_init(&dd->eep_lock);

	if (qib_mini_init)
		goto done;

	ret = init_pioavailregs(dd);
	init_shadow_tids(dd);

	qib_get_eeprom_info(dd);

	/* setup time (don't start yet) to verify we got interrupt */
	timer_setup(&dd->intrchk_timer, verify_interrupt, 0);
done:
	return ret;
}

/**
 * init_after_reset - re-initialize after a reset
 * @dd: the qlogic_ib device
 *
 * sanity check at least some of the values after reset, and
 * ensure no receive or transmit (explicitly, in case reset
 * failed
 */
static int init_after_reset(struct qib_devdata *dd)
{
	int i;

	/*
	 * Ensure chip does no sends or receives, tail updates, or
	 * pioavail updates while we re-initialize.  This is mostly
	 * for the driver data structures, not chip registers.
	 */
	for (i = 0; i < dd->num_pports; ++i) {
		/*
		 * ctxt == -1 means "all contexts". Only really safe for
		 * _dis_abling things, as here.
		 */
		dd->f_rcvctrl(dd->pport + i, QIB_RCVCTRL_CTXT_DIS |
				  QIB_RCVCTRL_INTRAVAIL_DIS |
				  QIB_RCVCTRL_TAILUPD_DIS, -1);
		/* Redundant across ports for some, but no big deal.  */
		dd->f_sendctrl(dd->pport + i, QIB_SENDCTRL_SEND_DIS |
			QIB_SENDCTRL_AVAIL_DIS);
	}

	return 0;
}

static void enable_chip(struct qib_devdata *dd)
{
	u64 rcvmask;
	int i;

	/*
	 * Enable PIO send, and update of PIOavail regs to memory.
	 */
	for (i = 0; i < dd->num_pports; ++i)
		dd->f_sendctrl(dd->pport + i, QIB_SENDCTRL_SEND_ENB |
			QIB_SENDCTRL_AVAIL_ENB);
	/*
	 * Enable kernel ctxts' receive and receive interrupt.
	 * Other ctxts done as user opens and inits them.
	 */
	rcvmask = QIB_RCVCTRL_CTXT_ENB | QIB_RCVCTRL_INTRAVAIL_ENB;
	rcvmask |= (dd->flags & QIB_NODMA_RTAIL) ?
		  QIB_RCVCTRL_TAILUPD_DIS : QIB_RCVCTRL_TAILUPD_ENB;
	for (i = 0; dd->rcd && i < dd->first_user_ctxt; ++i) {
		struct qib_ctxtdata *rcd = dd->rcd[i];

		if (rcd)
			dd->f_rcvctrl(rcd->ppd, rcvmask, i);
	}
}

static void verify_interrupt(struct timer_list *t)
{
	struct qib_devdata *dd = from_timer(dd, t, intrchk_timer);
	u64 int_counter;

	if (!dd)
		return; /* being torn down */

	/*
	 * If we don't have a lid or any interrupts, let the user know and
	 * don't bother checking again.
	 */
	int_counter = qib_int_counter(dd) - dd->z_int_counter;
	if (int_counter == 0) {
		if (!dd->f_intr_fallback(dd))
			dev_err(&dd->pcidev->dev,
				"No interrupts detected, not usable.\n");
		else /* re-arm the timer to see if fallback works */
			mod_timer(&dd->intrchk_timer, jiffies + HZ/2);
	}
}

static void init_piobuf_state(struct qib_devdata *dd)
{
	int i, pidx;
	u32 uctxts;

	/*
	 * Ensure all buffers are free, and fifos empty.  Buffers
	 * are common, so only do once for port 0.
	 *
	 * After enable and qib_chg_pioavailkernel so we can safely
	 * enable pioavail updates and PIOENABLE.  After this, packets
	 * are ready and able to go out.
	 */
	dd->f_sendctrl(dd->pport, QIB_SENDCTRL_DISARM_ALL);
	for (pidx = 0; pidx < dd->num_pports; ++pidx)
		dd->f_sendctrl(dd->pport + pidx, QIB_SENDCTRL_FLUSH);

	/*
	 * If not all sendbufs are used, add the one to each of the lower
	 * numbered contexts.  pbufsctxt and lastctxt_piobuf are
	 * calculated in chip-specific code because it may cause some
	 * chip-specific adjustments to be made.
	 */
	uctxts = dd->cfgctxts - dd->first_user_ctxt;
	dd->ctxts_extrabuf = dd->pbufsctxt ?
		dd->lastctxt_piobuf - (dd->pbufsctxt * uctxts) : 0;

	/*
	 * Set up the shadow copies of the piobufavail registers,
	 * which we compare against the chip registers for now, and
	 * the in memory DMA'ed copies of the registers.
	 * By now pioavail updates to memory should have occurred, so
	 * copy them into our working/shadow registers; this is in
	 * case something went wrong with abort, but mostly to get the
	 * initial values of the generation bit correct.
	 */
	for (i = 0; i < dd->pioavregs; i++) {
		__le64 tmp;

		tmp = dd->pioavailregs_dma[i];
		/*
		 * Don't need to worry about pioavailkernel here
		 * because we will call qib_chg_pioavailkernel() later
		 * in initialization, to busy out buffers as needed.
		 */
		dd->pioavailshadow[i] = le64_to_cpu(tmp);
	}
	while (i < ARRAY_SIZE(dd->pioavailshadow))
		dd->pioavailshadow[i++] = 0; /* for debugging sanity */

	/* after pioavailshadow is setup */
	qib_chg_pioavailkernel(dd, 0, dd->piobcnt2k + dd->piobcnt4k,
			       TXCHK_CHG_TYPE_KERN, NULL);
	dd->f_initvl15_bufs(dd);
}

/**
 * qib_create_workqueues - create per port workqueues
 * @dd: the qlogic_ib device
 */
static int qib_create_workqueues(struct qib_devdata *dd)
{
	int pidx;
	struct qib_pportdata *ppd;

	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
		ppd = dd->pport + pidx;
		if (!ppd->qib_wq) {
			char wq_name[8]; /* 3 + 2 + 1 + 1 + 1 */

			snprintf(wq_name, sizeof(wq_name), "qib%d_%d",
				dd->unit, pidx);
			ppd->qib_wq = alloc_ordered_workqueue(wq_name,
							      WQ_MEM_RECLAIM);
			if (!ppd->qib_wq)
				goto wq_error;
		}
	}
	return 0;
wq_error:
	pr_err("create_singlethread_workqueue failed for port %d\n",
		pidx + 1);
	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
		ppd = dd->pport + pidx;
		if (ppd->qib_wq) {
			destroy_workqueue(ppd->qib_wq);
			ppd->qib_wq = NULL;
		}
	}
	return -ENOMEM;
}

static void qib_free_pportdata(struct qib_pportdata *ppd)
{
	free_percpu(ppd->ibport_data.pmastats);
	free_percpu(ppd->ibport_data.rvp.rc_acks);
	free_percpu(ppd->ibport_data.rvp.rc_qacks);
	free_percpu(ppd->ibport_data.rvp.rc_delayed_comp);
	ppd->ibport_data.pmastats = NULL;
}

/**
 * qib_init - do the actual initialization sequence on the chip
 * @dd: the qlogic_ib device
 * @reinit: reinitializing, so don't allocate new memory
 *
 * Do the actual initialization sequence on the chip.  This is done
 * both from the init routine called from the PCI infrastructure, and
 * when we reset the chip, or detect that it was reset internally,
 * or it's administratively re-enabled.
 *
 * Memory allocation here and in called routines is only done in
 * the first case (reinit == 0).  We have to be careful, because even
 * without memory allocation, we need to re-write all the chip registers
 * TIDs, etc. after the reset or enable has completed.
 */
int qib_init(struct qib_devdata *dd, int reinit)
{
	int ret = 0, pidx, lastfail = 0;
	u32 portok = 0;
	unsigned i;
	struct qib_ctxtdata *rcd;
	struct qib_pportdata *ppd;
	unsigned long flags;

	/* Set linkstate to unknown, so we can watch for a transition. */
	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
		ppd = dd->pport + pidx;
		spin_lock_irqsave(&ppd->lflags_lock, flags);
		ppd->lflags &= ~(QIBL_LINKACTIVE | QIBL_LINKARMED |
				 QIBL_LINKDOWN | QIBL_LINKINIT |
				 QIBL_LINKV);
		spin_unlock_irqrestore(&ppd->lflags_lock, flags);
	}

	if (reinit)
		ret = init_after_reset(dd);
	else
		ret = loadtime_init(dd);
	if (ret)
		goto done;

	/* Bypass most chip-init, to get to device creation */
	if (qib_mini_init)
		return 0;

	ret = dd->f_late_initreg(dd);
	if (ret)
		goto done;

	/* dd->rcd can be NULL if early init failed */
	for (i = 0; dd->rcd && i < dd->first_user_ctxt; ++i) {
		/*
		 * Set up the (kernel) rcvhdr queue and egr TIDs.  If doing
		 * re-init, the simplest way to handle this is to free
		 * existing, and re-allocate.
		 * Need to re-create rest of ctxt 0 ctxtdata as well.
		 */
		rcd = dd->rcd[i];
		if (!rcd)
			continue;

		lastfail = qib_create_rcvhdrq(dd, rcd);
		if (!lastfail)
			lastfail = qib_setup_eagerbufs(rcd);
		if (lastfail)
			qib_dev_err(dd,
				"failed to allocate kernel ctxt's rcvhdrq and/or egr bufs\n");
	}

	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
		int mtu;

		if (lastfail)
			ret = lastfail;
		ppd = dd->pport + pidx;
		mtu = ib_mtu_enum_to_int(qib_ibmtu);
		if (mtu == -1) {
			mtu = QIB_DEFAULT_MTU;
			qib_ibmtu = 0; /* don't leave invalid value */
		}
		/* set max we can ever have for this driver load */
		ppd->init_ibmaxlen = min(mtu > 2048 ?
					 dd->piosize4k : dd->piosize2k,
					 dd->rcvegrbufsize +
					 (dd->rcvhdrentsize << 2));
		/*
		 * Have to initialize ibmaxlen, but this will normally
		 * change immediately in qib_set_mtu().
		 */
		ppd->ibmaxlen = ppd->init_ibmaxlen;
		qib_set_mtu(ppd, mtu);

		spin_lock_irqsave(&ppd->lflags_lock, flags);
		ppd->lflags |= QIBL_IB_LINK_DISABLED;
		spin_unlock_irqrestore(&ppd->lflags_lock, flags);

		lastfail = dd->f_bringup_serdes(ppd);
		if (lastfail) {
			qib_devinfo(dd->pcidev,
				 "Failed to bringup IB port %u\n", ppd->port);
			lastfail = -ENETDOWN;
			continue;
		}

		portok++;
	}

	if (!portok) {
		/* none of the ports initialized */
		if (!ret && lastfail)
			ret = lastfail;
		else if (!ret)
			ret = -ENETDOWN;
		/* but continue on, so we can debug cause */
	}

	enable_chip(dd);

	init_piobuf_state(dd);

done:
	if (!ret) {
		/* chip is OK for user apps; mark it as initialized */
		for (pidx = 0; pidx < dd->num_pports; ++pidx) {
			ppd = dd->pport + pidx;
			/*
			 * Set status even if port serdes is not initialized
			 * so that diags will work.
			 */
			*ppd->statusp |= QIB_STATUS_CHIP_PRESENT |
				QIB_STATUS_INITTED;
			if (!ppd->link_speed_enabled)
				continue;
			if (dd->flags & QIB_HAS_SEND_DMA)
				ret = qib_setup_sdma(ppd);
			timer_setup(&ppd->hol_timer, qib_hol_event, 0);
			ppd->hol_state = QIB_HOL_UP;
		}

		/* now we can enable all interrupts from the chip */
		dd->f_set_intr_state(dd, 1);

		/*
		 * Setup to verify we get an interrupt, and fallback
		 * to an alternate if necessary and possible.
		 */
		mod_timer(&dd->intrchk_timer, jiffies + HZ/2);
		/* start stats retrieval timer */
		mod_timer(&dd->stats_timer, jiffies + HZ * ACTIVITY_TIMER);
	}

	/* if ret is non-zero, we probably should do some cleanup here... */
	return ret;
}

/*
 * These next two routines are placeholders in case we don't have per-arch
 * code for controlling write combining.  If explicit control of write
 * combining is not available, performance will probably be awful.
 */

int __attribute__((weak)) qib_enable_wc(struct qib_devdata *dd)
{
	return -EOPNOTSUPP;
}

void __attribute__((weak)) qib_disable_wc(struct qib_devdata *dd)
{
}

struct qib_devdata *qib_lookup(int unit)
{
	return xa_load(&qib_dev_table, unit);
}

/*
 * Stop the timers during unit shutdown, or after an error late
 * in initialization.
 */
static void qib_stop_timers(struct qib_devdata *dd)
{
	struct qib_pportdata *ppd;
	int pidx;

	if (dd->stats_timer.function)
		del_timer_sync(&dd->stats_timer);
	if (dd->intrchk_timer.function)
		del_timer_sync(&dd->intrchk_timer);
	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
		ppd = dd->pport + pidx;
		if (ppd->hol_timer.function)
			del_timer_sync(&ppd->hol_timer);
		if (ppd->led_override_timer.function) {
			del_timer_sync(&ppd->led_override_timer);
			atomic_set(&ppd->led_override_timer_active, 0);
		}
		if (ppd->symerr_clear_timer.function)
			del_timer_sync(&ppd->symerr_clear_timer);
	}
}

/**
 * qib_shutdown_device - shut down a device
 * @dd: the qlogic_ib device
 *
 * This is called to make the device quiet when we are about to
 * unload the driver, and also when the device is administratively
 * disabled.   It does not free any data structures.
 * Everything it does has to be setup again by qib_init(dd, 1)
 */
static void qib_shutdown_device(struct qib_devdata *dd)
{
	struct qib_pportdata *ppd;
	unsigned pidx;

	if (dd->flags & QIB_SHUTDOWN)
		return;
	dd->flags |= QIB_SHUTDOWN;

	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
		ppd = dd->pport + pidx;

		spin_lock_irq(&ppd->lflags_lock);
		ppd->lflags &= ~(QIBL_LINKDOWN | QIBL_LINKINIT |
				 QIBL_LINKARMED | QIBL_LINKACTIVE |
				 QIBL_LINKV);
		spin_unlock_irq(&ppd->lflags_lock);
		*ppd->statusp &= ~(QIB_STATUS_IB_CONF | QIB_STATUS_IB_READY);
	}
	dd->flags &= ~QIB_INITTED;

	/* mask interrupts, but not errors */
	dd->f_set_intr_state(dd, 0);

	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
		ppd = dd->pport + pidx;
		dd->f_rcvctrl(ppd, QIB_RCVCTRL_TAILUPD_DIS |
				   QIB_RCVCTRL_CTXT_DIS |
				   QIB_RCVCTRL_INTRAVAIL_DIS |
				   QIB_RCVCTRL_PKEY_ENB, -1);
		/*
		 * Gracefully stop all sends allowing any in progress to
		 * trickle out first.
		 */
		dd->f_sendctrl(ppd, QIB_SENDCTRL_CLEAR);
	}

	/*
	 * Enough for anything that's going to trickle out to have actually
	 * done so.
	 */
	udelay(20);

	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
		ppd = dd->pport + pidx;
		dd->f_setextled(ppd, 0); /* make sure LEDs are off */

		if (dd->flags & QIB_HAS_SEND_DMA)
			qib_teardown_sdma(ppd);

		dd->f_sendctrl(ppd, QIB_SENDCTRL_AVAIL_DIS |
				    QIB_SENDCTRL_SEND_DIS);
		/*
		 * Clear SerdesEnable.
		 * We can't count on interrupts since we are stopping.
		 */
		dd->f_quiet_serdes(ppd);

		if (ppd->qib_wq) {
			destroy_workqueue(ppd->qib_wq);
			ppd->qib_wq = NULL;
		}
		qib_free_pportdata(ppd);
	}

}

/**
 * qib_free_ctxtdata - free a context's allocated data
 * @dd: the qlogic_ib device
 * @rcd: the ctxtdata structure
 *
 * free up any allocated data for a context
 * This should not touch anything that would affect a simultaneous
 * re-allocation of context data, because it is called after qib_mutex
 * is released (and can be called from reinit as well).
 * It should never change any chip state, or global driver state.
 */
void qib_free_ctxtdata(struct qib_devdata *dd, struct qib_ctxtdata *rcd)
{
	if (!rcd)
		return;

	if (rcd->rcvhdrq) {
		dma_free_coherent(&dd->pcidev->dev, rcd->rcvhdrq_size,
				  rcd->rcvhdrq, rcd->rcvhdrq_phys);
		rcd->rcvhdrq = NULL;
		if (rcd->rcvhdrtail_kvaddr) {
			dma_free_coherent(&dd->pcidev->dev, PAGE_SIZE,
					  rcd->rcvhdrtail_kvaddr,
					  rcd->rcvhdrqtailaddr_phys);
			rcd->rcvhdrtail_kvaddr = NULL;
		}
	}
	if (rcd->rcvegrbuf) {
		unsigned e;

		for (e = 0; e < rcd->rcvegrbuf_chunks; e++) {
			void *base = rcd->rcvegrbuf[e];
			size_t size = rcd->rcvegrbuf_size;

			dma_free_coherent(&dd->pcidev->dev, size,
					  base, rcd->rcvegrbuf_phys[e]);
		}
		kfree(rcd->rcvegrbuf);
		rcd->rcvegrbuf = NULL;
		kfree(rcd->rcvegrbuf_phys);
		rcd->rcvegrbuf_phys = NULL;
		rcd->rcvegrbuf_chunks = 0;
	}

	kfree(rcd->tid_pg_list);
	vfree(rcd->user_event_mask);
	vfree(rcd->subctxt_uregbase);
	vfree(rcd->subctxt_rcvegrbuf);
	vfree(rcd->subctxt_rcvhdr_base);
#ifdef CONFIG_DEBUG_FS
	kfree(rcd->opstats);
	rcd->opstats = NULL;
#endif
	kfree(rcd);
}

/*
 * Perform a PIO buffer bandwidth write test, to verify proper system
 * configuration.  Even when all the setup calls work, occasionally
 * BIOS or other issues can prevent write combining from working, or
 * can cause other bandwidth problems to the chip.
 *
 * This test simply writes the same buffer over and over again, and
 * measures close to the peak bandwidth to the chip (not testing
 * data bandwidth to the wire).   On chips that use an address-based
 * trigger to send packets to the wire, this is easy.  On chips that
 * use a count to trigger, we want to make sure that the packet doesn't
 * go out on the wire, or trigger flow control checks.
 */
static void qib_verify_pioperf(struct qib_devdata *dd)
{
	u32 pbnum, cnt, lcnt;
	u32 __iomem *piobuf;
	u32 *addr;
	u64 msecs, emsecs;

	piobuf = dd->f_getsendbuf(dd->pport, 0ULL, &pbnum);
	if (!piobuf) {
		qib_devinfo(dd->pcidev,
			 "No PIObufs for checking perf, skipping\n");
		return;
	}

	/*
	 * Enough to give us a reasonable test, less than piobuf size, and
	 * likely multiple of store buffer length.
	 */
	cnt = 1024;

	addr = vmalloc(cnt);
	if (!addr)
		goto done;

	preempt_disable();  /* we want reasonably accurate elapsed time */
	msecs = 1 + jiffies_to_msecs(jiffies);
	for (lcnt = 0; lcnt < 10000U; lcnt++) {
		/* wait until we cross msec boundary */
		if (jiffies_to_msecs(jiffies) >= msecs)
			break;
		udelay(1);
	}

	dd->f_set_armlaunch(dd, 0);

	/*
	 * length 0, no dwords actually sent
	 */
	writeq(0, piobuf);
	qib_flush_wc();

	/*
	 * This is only roughly accurate, since even with preempt we
	 * still take interrupts that could take a while.   Running for
	 * >= 5 msec seems to get us "close enough" to accurate values.
	 */
	msecs = jiffies_to_msecs(jiffies);
	for (emsecs = lcnt = 0; emsecs <= 5UL; lcnt++) {
		qib_pio_copy(piobuf + 64, addr, cnt >> 2);
		emsecs = jiffies_to_msecs(jiffies) - msecs;
	}

	/* 1 GiB/sec, slightly over IB SDR line rate */
	if (lcnt < (emsecs * 1024U))
		qib_dev_err(dd,
			    "Performance problem: bandwidth to PIO buffers is only %u MiB/sec\n",
			    lcnt / (u32) emsecs);

	preempt_enable();

	vfree(addr);

done:
	/* disarm piobuf, so it's available again */
	dd->f_sendctrl(dd->pport, QIB_SENDCTRL_DISARM_BUF(pbnum));
	qib_sendbuf_done(dd, pbnum);
	dd->f_set_armlaunch(dd, 1);
}

void qib_free_devdata(struct qib_devdata *dd)
{
	unsigned long flags;

	xa_lock_irqsave(&qib_dev_table, flags);
	__xa_erase(&qib_dev_table, dd->unit);
	xa_unlock_irqrestore(&qib_dev_table, flags);

#ifdef CONFIG_DEBUG_FS
	qib_dbg_ibdev_exit(&dd->verbs_dev);
#endif
	free_percpu(dd->int_counter);
	rvt_dealloc_device(&dd->verbs_dev.rdi);
}

u64 qib_int_counter(struct qib_devdata *dd)
{
	int cpu;
	u64 int_counter = 0;

	for_each_possible_cpu(cpu)
		int_counter += *per_cpu_ptr(dd->int_counter, cpu);
	return int_counter;
}

u64 qib_sps_ints(void)
{
	unsigned long index, flags;
	struct qib_devdata *dd;
	u64 sps_ints = 0;

	xa_lock_irqsave(&qib_dev_table, flags);
	xa_for_each(&qib_dev_table, index, dd) {
		sps_ints += qib_int_counter(dd);
	}
	xa_unlock_irqrestore(&qib_dev_table, flags);
	return sps_ints;
}

/*
 * Allocate our primary per-unit data structure.  Must be done via verbs
 * allocator, because the verbs cleanup process both does cleanup and
 * free of the data structure.
 * "extra" is for chip-specific data.
 */
struct qib_devdata *qib_alloc_devdata(struct pci_dev *pdev, size_t extra)
{
	struct qib_devdata *dd;
	int ret, nports;

	/* extra is * number of ports */
	nports = extra / sizeof(struct qib_pportdata);
	dd = (struct qib_devdata *)rvt_alloc_device(sizeof(*dd) + extra,
						    nports);
	if (!dd)
		return ERR_PTR(-ENOMEM);

	ret = xa_alloc_irq(&qib_dev_table, &dd->unit, dd, xa_limit_32b,
			GFP_KERNEL);
	if (ret < 0) {
		qib_early_err(&pdev->dev,
			      "Could not allocate unit ID: error %d\n", -ret);
		goto bail;
	}
	rvt_set_ibdev_name(&dd->verbs_dev.rdi, "%s%d", "qib", dd->unit);

	dd->int_counter = alloc_percpu(u64);
	if (!dd->int_counter) {
		ret = -ENOMEM;
		qib_early_err(&pdev->dev,
			      "Could not allocate per-cpu int_counter\n");
		goto bail;
	}

	if (!qib_cpulist_count) {
		u32 count = num_online_cpus();

		qib_cpulist = kcalloc(BITS_TO_LONGS(count), sizeof(long),
				      GFP_KERNEL);
		if (qib_cpulist)
			qib_cpulist_count = count;
	}
#ifdef CONFIG_DEBUG_FS
	qib_dbg_ibdev_init(&dd->verbs_dev);
#endif
	return dd;
bail:
	if (!list_empty(&dd->list))
		list_del_init(&dd->list);
	rvt_dealloc_device(&dd->verbs_dev.rdi);
	return ERR_PTR(ret);
}

/*
 * Called from freeze mode handlers, and from PCI error
 * reporting code.  Should be paranoid about state of
 * system and data structures.
 */
void qib_disable_after_error(struct qib_devdata *dd)
{
	if (dd->flags & QIB_INITTED) {
		u32 pidx;

		dd->flags &= ~QIB_INITTED;
		if (dd->pport)
			for (pidx = 0; pidx < dd->num_pports; ++pidx) {
				struct qib_pportdata *ppd;

				ppd = dd->pport + pidx;
				if (dd->flags & QIB_PRESENT) {
					qib_set_linkstate(ppd,
						QIB_IB_LINKDOWN_DISABLE);
					dd->f_setextled(ppd, 0);
				}
				*ppd->statusp &= ~QIB_STATUS_IB_READY;
			}
	}

	/*
	 * Mark as having had an error for driver, and also
	 * for /sys and status word mapped to user programs.
	 * This marks unit as not usable, until reset.
	 */
	if (dd->devstatusp)
		*dd->devstatusp |= QIB_STATUS_HWERROR;
}

static void qib_remove_one(struct pci_dev *);
static int qib_init_one(struct pci_dev *, const struct pci_device_id *);
static void qib_shutdown_one(struct pci_dev *);

#define DRIVER_LOAD_MSG "Intel " QIB_DRV_NAME " loaded: "
#define PFX QIB_DRV_NAME ": "

static const struct pci_device_id qib_pci_tbl[] = {
	{ PCI_DEVICE(PCI_VENDOR_ID_PATHSCALE, PCI_DEVICE_ID_QLOGIC_IB_6120) },
	{ PCI_DEVICE(PCI_VENDOR_ID_QLOGIC, PCI_DEVICE_ID_QLOGIC_IB_7220) },
	{ PCI_DEVICE(PCI_VENDOR_ID_QLOGIC, PCI_DEVICE_ID_QLOGIC_IB_7322) },
	{ 0, }
};

MODULE_DEVICE_TABLE(pci, qib_pci_tbl);

static struct pci_driver qib_driver = {
	.name = QIB_DRV_NAME,
	.probe = qib_init_one,
	.remove = qib_remove_one,
	.shutdown = qib_shutdown_one,
	.id_table = qib_pci_tbl,
	.err_handler = &qib_pci_err_handler,
};

#ifdef CONFIG_INFINIBAND_QIB_DCA

static int qib_notify_dca(struct notifier_block *, unsigned long, void *);
static struct notifier_block dca_notifier = {
	.notifier_call  = qib_notify_dca,
	.next           = NULL,
	.priority       = 0
};

static int qib_notify_dca_device(struct device *device, void *data)
{
	struct qib_devdata *dd = dev_get_drvdata(device);
	unsigned long event = *(unsigned long *)data;

	return dd->f_notify_dca(dd, event);
}

static int qib_notify_dca(struct notifier_block *nb, unsigned long event,
					  void *p)
{
	int rval;

	rval = driver_for_each_device(&qib_driver.driver, NULL,
				      &event, qib_notify_dca_device);
	return rval ? NOTIFY_BAD : NOTIFY_DONE;
}

#endif

/*
 * Do all the generic driver unit- and chip-independent memory
 * allocation and initialization.
 */
static int __init qib_ib_init(void)
{
	int ret;

	ret = qib_dev_init();
	if (ret)
		goto bail;

	/*
	 * These must be called before the driver is registered with
	 * the PCI subsystem.
	 */
#ifdef CONFIG_INFINIBAND_QIB_DCA
	dca_register_notify(&dca_notifier);
#endif
#ifdef CONFIG_DEBUG_FS
	qib_dbg_init();
#endif
	ret = pci_register_driver(&qib_driver);
	if (ret < 0) {
		pr_err("Unable to register driver: error %d\n", -ret);
		goto bail_dev;
	}

	/* not fatal if it doesn't work */
	if (qib_init_qibfs())
		pr_err("Unable to register ipathfs\n");
	goto bail; /* all OK */

bail_dev:
#ifdef CONFIG_INFINIBAND_QIB_DCA
	dca_unregister_notify(&dca_notifier);
#endif
#ifdef CONFIG_DEBUG_FS
	qib_dbg_exit();
#endif
	qib_dev_cleanup();
bail:
	return ret;
}

module_init(qib_ib_init);

/*
 * Do the non-unit driver cleanup, memory free, etc. at unload.
 */
static void __exit qib_ib_cleanup(void)
{
	int ret;

	ret = qib_exit_qibfs();
	if (ret)
		pr_err(
			"Unable to cleanup counter filesystem: error %d\n",
			-ret);

#ifdef CONFIG_INFINIBAND_QIB_DCA
	dca_unregister_notify(&dca_notifier);
#endif
	pci_unregister_driver(&qib_driver);
#ifdef CONFIG_DEBUG_FS
	qib_dbg_exit();
#endif

	qib_cpulist_count = 0;
	kfree(qib_cpulist);

	WARN_ON(!xa_empty(&qib_dev_table));
	qib_dev_cleanup();
}

module_exit(qib_ib_cleanup);

/* this can only be called after a successful initialization */
static void cleanup_device_data(struct qib_devdata *dd)
{
	int ctxt;
	int pidx;
	struct qib_ctxtdata **tmp;
	unsigned long flags;

	/* users can't do anything more with chip */
	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
		if (dd->pport[pidx].statusp)
			*dd->pport[pidx].statusp &= ~QIB_STATUS_CHIP_PRESENT;

		spin_lock(&dd->pport[pidx].cc_shadow_lock);

		kfree(dd->pport[pidx].congestion_entries);
		dd->pport[pidx].congestion_entries = NULL;
		kfree(dd->pport[pidx].ccti_entries);
		dd->pport[pidx].ccti_entries = NULL;
		kfree(dd->pport[pidx].ccti_entries_shadow);
		dd->pport[pidx].ccti_entries_shadow = NULL;
		kfree(dd->pport[pidx].congestion_entries_shadow);
		dd->pport[pidx].congestion_entries_shadow = NULL;

		spin_unlock(&dd->pport[pidx].cc_shadow_lock);
	}

	qib_disable_wc(dd);

	if (dd->pioavailregs_dma) {
		dma_free_coherent(&dd->pcidev->dev, PAGE_SIZE,
				  (void *) dd->pioavailregs_dma,
				  dd->pioavailregs_phys);
		dd->pioavailregs_dma = NULL;
	}

	if (dd->pageshadow) {
		struct page **tmpp = dd->pageshadow;
		dma_addr_t *tmpd = dd->physshadow;
		int i;

		for (ctxt = 0; ctxt < dd->cfgctxts; ctxt++) {
			int ctxt_tidbase = ctxt * dd->rcvtidcnt;
			int maxtid = ctxt_tidbase + dd->rcvtidcnt;

			for (i = ctxt_tidbase; i < maxtid; i++) {
				if (!tmpp[i])
					continue;
				pci_unmap_page(dd->pcidev, tmpd[i],
					       PAGE_SIZE, PCI_DMA_FROMDEVICE);
				qib_release_user_pages(&tmpp[i], 1);
				tmpp[i] = NULL;
			}
		}

		dd->pageshadow = NULL;
		vfree(tmpp);
		dd->physshadow = NULL;
		vfree(tmpd);
	}

	/*
	 * Free any resources still in use (usually just kernel contexts)
	 * at unload; we do for ctxtcnt, because that's what we allocate.
	 * We acquire lock to be really paranoid that rcd isn't being
	 * accessed from some interrupt-related code (that should not happen,
	 * but best to be sure).
	 */
	spin_lock_irqsave(&dd->uctxt_lock, flags);
	tmp = dd->rcd;
	dd->rcd = NULL;
	spin_unlock_irqrestore(&dd->uctxt_lock, flags);
	for (ctxt = 0; tmp && ctxt < dd->ctxtcnt; ctxt++) {
		struct qib_ctxtdata *rcd = tmp[ctxt];

		tmp[ctxt] = NULL; /* debugging paranoia */
		qib_free_ctxtdata(dd, rcd);
	}
	kfree(tmp);
}

/*
 * Clean up on unit shutdown, or error during unit load after
 * successful initialization.
 */
static void qib_postinit_cleanup(struct qib_devdata *dd)
{
	/*
	 * Clean up chip-specific stuff.
	 * We check for NULL here, because it's outside
	 * the kregbase check, and we need to call it
	 * after the free_irq.  Thus it's possible that
	 * the function pointers were never initialized.
	 */
	if (dd->f_cleanup)
		dd->f_cleanup(dd);

	qib_pcie_ddcleanup(dd);

	cleanup_device_data(dd);

	qib_free_devdata(dd);
}

static int qib_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
{
	int ret, j, pidx, initfail;
	struct qib_devdata *dd = NULL;

	ret = qib_pcie_init(pdev, ent);
	if (ret)
		goto bail;

	/*
	 * Do device-specific initialiation, function table setup, dd
	 * allocation, etc.
	 */
	switch (ent->device) {
	case PCI_DEVICE_ID_QLOGIC_IB_6120:
#ifdef CONFIG_PCI_MSI
		dd = qib_init_iba6120_funcs(pdev, ent);
#else
		qib_early_err(&pdev->dev,
			"Intel PCIE device 0x%x cannot work if CONFIG_PCI_MSI is not enabled\n",
			ent->device);
		dd = ERR_PTR(-ENODEV);
#endif
		break;

	case PCI_DEVICE_ID_QLOGIC_IB_7220:
		dd = qib_init_iba7220_funcs(pdev, ent);
		break;

	case PCI_DEVICE_ID_QLOGIC_IB_7322:
		dd = qib_init_iba7322_funcs(pdev, ent);
		break;

	default:
		qib_early_err(&pdev->dev,
			"Failing on unknown Intel deviceid 0x%x\n",
			ent->device);
		ret = -ENODEV;
	}

	if (IS_ERR(dd))
		ret = PTR_ERR(dd);
	if (ret)
		goto bail; /* error already printed */

	ret = qib_create_workqueues(dd);
	if (ret)
		goto bail;

	/* do the generic initialization */
	initfail = qib_init(dd, 0);

	ret = qib_register_ib_device(dd);

	/*
	 * Now ready for use.  this should be cleared whenever we
	 * detect a reset, or initiate one.  If earlier failure,
	 * we still create devices, so diags, etc. can be used
	 * to determine cause of problem.
	 */
	if (!qib_mini_init && !initfail && !ret)
		dd->flags |= QIB_INITTED;

	j = qib_device_create(dd);
	if (j)
		qib_dev_err(dd, "Failed to create /dev devices: %d\n", -j);
	j = qibfs_add(dd);
	if (j)
		qib_dev_err(dd, "Failed filesystem setup for counters: %d\n",
			    -j);

	if (qib_mini_init || initfail || ret) {
		qib_stop_timers(dd);
		flush_workqueue(ib_wq);
		for (pidx = 0; pidx < dd->num_pports; ++pidx)
			dd->f_quiet_serdes(dd->pport + pidx);
		if (qib_mini_init)
			goto bail;
		if (!j) {
			(void) qibfs_remove(dd);
			qib_device_remove(dd);
		}
		if (!ret)
			qib_unregister_ib_device(dd);
		qib_postinit_cleanup(dd);
		if (initfail)
			ret = initfail;
		goto bail;
	}

	ret = qib_enable_wc(dd);
	if (ret) {
		qib_dev_err(dd,
			"Write combining not enabled (err %d): performance may be poor\n",
			-ret);
		ret = 0;
	}

	qib_verify_pioperf(dd);
bail:
	return ret;
}

static void qib_remove_one(struct pci_dev *pdev)
{
	struct qib_devdata *dd = pci_get_drvdata(pdev);
	int ret;

	/* unregister from IB core */
	qib_unregister_ib_device(dd);

	/*
	 * Disable the IB link, disable interrupts on the device,
	 * clear dma engines, etc.
	 */
	if (!qib_mini_init)
		qib_shutdown_device(dd);

	qib_stop_timers(dd);

	/* wait until all of our (qsfp) queue_work() calls complete */
	flush_workqueue(ib_wq);

	ret = qibfs_remove(dd);
	if (ret)
		qib_dev_err(dd, "Failed counters filesystem cleanup: %d\n",
			    -ret);

	qib_device_remove(dd);

	qib_postinit_cleanup(dd);
}

static void qib_shutdown_one(struct pci_dev *pdev)
{
	struct qib_devdata *dd = pci_get_drvdata(pdev);

	qib_shutdown_device(dd);
}

/**
 * qib_create_rcvhdrq - create a receive header queue
 * @dd: the qlogic_ib device
 * @rcd: the context data
 *
 * This must be contiguous memory (from an i/o perspective), and must be
 * DMA'able (which means for some systems, it will go through an IOMMU,
 * or be forced into a low address range).
 */
int qib_create_rcvhdrq(struct qib_devdata *dd, struct qib_ctxtdata *rcd)
{
	unsigned amt;
	int old_node_id;

	if (!rcd->rcvhdrq) {
		dma_addr_t phys_hdrqtail;
		gfp_t gfp_flags;

		amt = ALIGN(dd->rcvhdrcnt * dd->rcvhdrentsize *
			    sizeof(u32), PAGE_SIZE);
		gfp_flags = (rcd->ctxt >= dd->first_user_ctxt) ?
			GFP_USER : GFP_KERNEL;

		old_node_id = dev_to_node(&dd->pcidev->dev);
		set_dev_node(&dd->pcidev->dev, rcd->node_id);
		rcd->rcvhdrq = dma_alloc_coherent(
			&dd->pcidev->dev, amt, &rcd->rcvhdrq_phys,
			gfp_flags | __GFP_COMP);
		set_dev_node(&dd->pcidev->dev, old_node_id);

		if (!rcd->rcvhdrq) {
			qib_dev_err(dd,
				"attempt to allocate %d bytes for ctxt %u rcvhdrq failed\n",
				amt, rcd->ctxt);
			goto bail;
		}

		if (rcd->ctxt >= dd->first_user_ctxt) {
			rcd->user_event_mask = vmalloc_user(PAGE_SIZE);
			if (!rcd->user_event_mask)
				goto bail_free_hdrq;
		}

		if (!(dd->flags & QIB_NODMA_RTAIL)) {
			set_dev_node(&dd->pcidev->dev, rcd->node_id);
			rcd->rcvhdrtail_kvaddr = dma_alloc_coherent(
				&dd->pcidev->dev, PAGE_SIZE, &phys_hdrqtail,
				gfp_flags);
			set_dev_node(&dd->pcidev->dev, old_node_id);
			if (!rcd->rcvhdrtail_kvaddr)
				goto bail_free;
			rcd->rcvhdrqtailaddr_phys = phys_hdrqtail;
		}

		rcd->rcvhdrq_size = amt;
	}

	/* clear for security and sanity on each use */
	memset(rcd->rcvhdrq, 0, rcd->rcvhdrq_size);
	if (rcd->rcvhdrtail_kvaddr)
		memset(rcd->rcvhdrtail_kvaddr, 0, PAGE_SIZE);
	return 0;

bail_free:
	qib_dev_err(dd,
		"attempt to allocate 1 page for ctxt %u rcvhdrqtailaddr failed\n",
		rcd->ctxt);
	vfree(rcd->user_event_mask);
	rcd->user_event_mask = NULL;
bail_free_hdrq:
	dma_free_coherent(&dd->pcidev->dev, amt, rcd->rcvhdrq,
			  rcd->rcvhdrq_phys);
	rcd->rcvhdrq = NULL;
bail:
	return -ENOMEM;
}

/**
 * qib_setup_eagerbufs - allocate eager buffers, both kernel and user contexts.
 * @rcd: the context we are setting up.
 *
 * Allocate the eager TID buffers and program them into hip.
 * They are no longer completely contiguous, we do multiple allocation
 * calls.  Otherwise we get the OOM code involved, by asking for too
 * much per call, with disastrous results on some kernels.
 */
int qib_setup_eagerbufs(struct qib_ctxtdata *rcd)
{
	struct qib_devdata *dd = rcd->dd;
	unsigned e, egrcnt, egrperchunk, chunk, egrsize, egroff;
	size_t size;
	gfp_t gfp_flags;
	int old_node_id;

	/*
	 * GFP_USER, but without GFP_FS, so buffer cache can be
	 * coalesced (we hope); otherwise, even at order 4,
	 * heavy filesystem activity makes these fail, and we can
	 * use compound pages.
	 */
	gfp_flags = __GFP_RECLAIM | __GFP_IO | __GFP_COMP;

	egrcnt = rcd->rcvegrcnt;
	egroff = rcd->rcvegr_tid_base;
	egrsize = dd->rcvegrbufsize;

	chunk = rcd->rcvegrbuf_chunks;
	egrperchunk = rcd->rcvegrbufs_perchunk;
	size = rcd->rcvegrbuf_size;
	if (!rcd->rcvegrbuf) {
		rcd->rcvegrbuf =
			kcalloc_node(chunk, sizeof(rcd->rcvegrbuf[0]),
				     GFP_KERNEL, rcd->node_id);
		if (!rcd->rcvegrbuf)
			goto bail;
	}
	if (!rcd->rcvegrbuf_phys) {
		rcd->rcvegrbuf_phys =
			kmalloc_array_node(chunk,
					   sizeof(rcd->rcvegrbuf_phys[0]),
					   GFP_KERNEL, rcd->node_id);
		if (!rcd->rcvegrbuf_phys)
			goto bail_rcvegrbuf;
	}
	for (e = 0; e < rcd->rcvegrbuf_chunks; e++) {
		if (rcd->rcvegrbuf[e])
			continue;

		old_node_id = dev_to_node(&dd->pcidev->dev);
		set_dev_node(&dd->pcidev->dev, rcd->node_id);
		rcd->rcvegrbuf[e] =
			dma_alloc_coherent(&dd->pcidev->dev, size,
					   &rcd->rcvegrbuf_phys[e],
					   gfp_flags);
		set_dev_node(&dd->pcidev->dev, old_node_id);
		if (!rcd->rcvegrbuf[e])
			goto bail_rcvegrbuf_phys;
	}

	rcd->rcvegr_phys = rcd->rcvegrbuf_phys[0];

	for (e = chunk = 0; chunk < rcd->rcvegrbuf_chunks; chunk++) {
		dma_addr_t pa = rcd->rcvegrbuf_phys[chunk];
		unsigned i;

		/* clear for security and sanity on each use */
		memset(rcd->rcvegrbuf[chunk], 0, size);

		for (i = 0; e < egrcnt && i < egrperchunk; e++, i++) {
			dd->f_put_tid(dd, e + egroff +
					  (u64 __iomem *)
					  ((char __iomem *)
					   dd->kregbase +
					   dd->rcvegrbase),
					  RCVHQ_RCV_TYPE_EAGER, pa);
			pa += egrsize;
		}
		cond_resched(); /* don't hog the cpu */
	}

	return 0;

bail_rcvegrbuf_phys:
	for (e = 0; e < rcd->rcvegrbuf_chunks && rcd->rcvegrbuf[e]; e++)
		dma_free_coherent(&dd->pcidev->dev, size,
				  rcd->rcvegrbuf[e], rcd->rcvegrbuf_phys[e]);
	kfree(rcd->rcvegrbuf_phys);
	rcd->rcvegrbuf_phys = NULL;
bail_rcvegrbuf:
	kfree(rcd->rcvegrbuf);
	rcd->rcvegrbuf = NULL;
bail:
	return -ENOMEM;
}

/*
 * Note: Changes to this routine should be mirrored
 * for the diagnostics routine qib_remap_ioaddr32().
 * There is also related code for VL15 buffers in qib_init_7322_variables().
 * The teardown code that unmaps is in qib_pcie_ddcleanup()
 */
int init_chip_wc_pat(struct qib_devdata *dd, u32 vl15buflen)
{
	u64 __iomem *qib_kregbase = NULL;
	void __iomem *qib_piobase = NULL;
	u64 __iomem *qib_userbase = NULL;
	u64 qib_kreglen;
	u64 qib_pio2koffset = dd->piobufbase & 0xffffffff;
	u64 qib_pio4koffset = dd->piobufbase >> 32;
	u64 qib_pio2klen = dd->piobcnt2k * dd->palign;
	u64 qib_pio4klen = dd->piobcnt4k * dd->align4k;
	u64 qib_physaddr = dd->physaddr;
	u64 qib_piolen;
	u64 qib_userlen = 0;

	/*
	 * Free the old mapping because the kernel will try to reuse the
	 * old mapping and not create a new mapping with the
	 * write combining attribute.
	 */
	iounmap(dd->kregbase);
	dd->kregbase = NULL;

	/*
	 * Assumes chip address space looks like:
	 *	- kregs + sregs + cregs + uregs (in any order)
	 *	- piobufs (2K and 4K bufs in either order)
	 * or:
	 *	- kregs + sregs + cregs (in any order)
	 *	- piobufs (2K and 4K bufs in either order)
	 *	- uregs
	 */
	if (dd->piobcnt4k == 0) {
		qib_kreglen = qib_pio2koffset;
		qib_piolen = qib_pio2klen;
	} else if (qib_pio2koffset < qib_pio4koffset) {
		qib_kreglen = qib_pio2koffset;
		qib_piolen = qib_pio4koffset + qib_pio4klen - qib_kreglen;
	} else {
		qib_kreglen = qib_pio4koffset;
		qib_piolen = qib_pio2koffset + qib_pio2klen - qib_kreglen;
	}
	qib_piolen += vl15buflen;
	/* Map just the configured ports (not all hw ports) */
	if (dd->uregbase > qib_kreglen)
		qib_userlen = dd->ureg_align * dd->cfgctxts;

	/* Sanity checks passed, now create the new mappings */
	qib_kregbase = ioremap(qib_physaddr, qib_kreglen);
	if (!qib_kregbase)
		goto bail;

	qib_piobase = ioremap_wc(qib_physaddr + qib_kreglen, qib_piolen);
	if (!qib_piobase)
		goto bail_kregbase;

	if (qib_userlen) {
		qib_userbase = ioremap(qib_physaddr + dd->uregbase,
					       qib_userlen);
		if (!qib_userbase)
			goto bail_piobase;
	}

	dd->kregbase = qib_kregbase;
	dd->kregend = (u64 __iomem *)
		((char __iomem *) qib_kregbase + qib_kreglen);
	dd->piobase = qib_piobase;
	dd->pio2kbase = (void __iomem *)
		(((char __iomem *) dd->piobase) +
		 qib_pio2koffset - qib_kreglen);
	if (dd->piobcnt4k)
		dd->pio4kbase = (void __iomem *)
			(((char __iomem *) dd->piobase) +
			 qib_pio4koffset - qib_kreglen);
	if (qib_userlen)
		/* ureg will now be accessed relative to dd->userbase */
		dd->userbase = qib_userbase;
	return 0;

bail_piobase:
	iounmap(qib_piobase);
bail_kregbase:
	iounmap(qib_kregbase);
bail:
	return -ENOMEM;
}