aboutsummaryrefslogtreecommitdiff
path: root/drivers/infiniband/hw/ipath/ipath_driver.c
blob: eb745234d8c3471db47494de29bd9bc24d615cfe (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
/*
 * Copyright (c) 2006 QLogic, Inc. 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/spinlock.h>
#include <linux/idr.h>
#include <linux/pci.h>
#include <linux/delay.h>
#include <linux/netdevice.h>
#include <linux/vmalloc.h>

#include "ipath_kernel.h"
#include "ips_common.h"
#include "ipath_layer.h"

static void ipath_update_pio_bufs(struct ipath_devdata *);

const char *ipath_get_unit_name(int unit)
{
	static char iname[16];
	snprintf(iname, sizeof iname, "infinipath%u", unit);
	return iname;
}

EXPORT_SYMBOL_GPL(ipath_get_unit_name);

#define DRIVER_LOAD_MSG "QLogic " IPATH_DRV_NAME " loaded: "
#define PFX IPATH_DRV_NAME ": "

/*
 * The size has to be longer than this string, so we can append
 * board/chip information to it in the init code.
 */
const char ipath_core_version[] = IPATH_IDSTR "\n";

static struct idr unit_table;
DEFINE_SPINLOCK(ipath_devs_lock);
LIST_HEAD(ipath_dev_list);

wait_queue_head_t ipath_sma_state_wait;

unsigned ipath_debug = __IPATH_INFO;

module_param_named(debug, ipath_debug, uint, S_IWUSR | S_IRUGO);
MODULE_PARM_DESC(debug, "mask for debug prints");
EXPORT_SYMBOL_GPL(ipath_debug);

MODULE_LICENSE("GPL");
MODULE_AUTHOR("QLogic <support@pathscale.com>");
MODULE_DESCRIPTION("QLogic InfiniPath driver");

const char *ipath_ibcstatus_str[] = {
	"Disabled",
	"LinkUp",
	"PollActive",
	"PollQuiet",
	"SleepDelay",
	"SleepQuiet",
	"LState6",		/* unused */
	"LState7",		/* unused */
	"CfgDebounce",
	"CfgRcvfCfg",
	"CfgWaitRmt",
	"CfgIdle",
	"RecovRetrain",
	"LState0xD",		/* unused */
	"RecovWaitRmt",
	"RecovIdle",
};

/*
 * These variables are initialized in the chip-specific files
 * but are defined here.
 */
u16 ipath_gpio_sda_num, ipath_gpio_scl_num;
u64 ipath_gpio_sda, ipath_gpio_scl;
u64 infinipath_i_bitsextant;
ipath_err_t infinipath_e_bitsextant, infinipath_hwe_bitsextant;
u32 infinipath_i_rcvavail_mask, infinipath_i_rcvurg_mask;

static void __devexit ipath_remove_one(struct pci_dev *);
static int __devinit ipath_init_one(struct pci_dev *,
				    const struct pci_device_id *);

/* Only needed for registration, nothing else needs this info */
#define PCI_VENDOR_ID_PATHSCALE 0x1fc1
#define PCI_DEVICE_ID_INFINIPATH_HT 0xd
#define PCI_DEVICE_ID_INFINIPATH_PE800 0x10

static const struct pci_device_id ipath_pci_tbl[] = {
	{ PCI_DEVICE(PCI_VENDOR_ID_PATHSCALE, PCI_DEVICE_ID_INFINIPATH_HT) },
	{ PCI_DEVICE(PCI_VENDOR_ID_PATHSCALE, PCI_DEVICE_ID_INFINIPATH_PE800) },
	{ 0, }
};

MODULE_DEVICE_TABLE(pci, ipath_pci_tbl);

static struct pci_driver ipath_driver = {
	.name = IPATH_DRV_NAME,
	.probe = ipath_init_one,
	.remove = __devexit_p(ipath_remove_one),
	.id_table = ipath_pci_tbl,
};


static inline void read_bars(struct ipath_devdata *dd, struct pci_dev *dev,
			     u32 *bar0, u32 *bar1)
{
	int ret;

	ret = pci_read_config_dword(dev, PCI_BASE_ADDRESS_0, bar0);
	if (ret)
		ipath_dev_err(dd, "failed to read bar0 before enable: "
			      "error %d\n", -ret);

	ret = pci_read_config_dword(dev, PCI_BASE_ADDRESS_1, bar1);
	if (ret)
		ipath_dev_err(dd, "failed to read bar1 before enable: "
			      "error %d\n", -ret);

	ipath_dbg("Read bar0 %x bar1 %x\n", *bar0, *bar1);
}

static void ipath_free_devdata(struct pci_dev *pdev,
			       struct ipath_devdata *dd)
{
	unsigned long flags;

	pci_set_drvdata(pdev, NULL);

	if (dd->ipath_unit != -1) {
		spin_lock_irqsave(&ipath_devs_lock, flags);
		idr_remove(&unit_table, dd->ipath_unit);
		list_del(&dd->ipath_list);
		spin_unlock_irqrestore(&ipath_devs_lock, flags);
	}
	vfree(dd);
}

static struct ipath_devdata *ipath_alloc_devdata(struct pci_dev *pdev)
{
	unsigned long flags;
	struct ipath_devdata *dd;
	int ret;

	if (!idr_pre_get(&unit_table, GFP_KERNEL)) {
		dd = ERR_PTR(-ENOMEM);
		goto bail;
	}

	dd = vmalloc(sizeof(*dd));
	if (!dd) {
		dd = ERR_PTR(-ENOMEM);
		goto bail;
	}
	memset(dd, 0, sizeof(*dd));
	dd->ipath_unit = -1;

	spin_lock_irqsave(&ipath_devs_lock, flags);

	ret = idr_get_new(&unit_table, dd, &dd->ipath_unit);
	if (ret < 0) {
		printk(KERN_ERR IPATH_DRV_NAME
		       ": Could not allocate unit ID: error %d\n", -ret);
		ipath_free_devdata(pdev, dd);
		dd = ERR_PTR(ret);
		goto bail_unlock;
	}

	dd->pcidev = pdev;
	pci_set_drvdata(pdev, dd);

	list_add(&dd->ipath_list, &ipath_dev_list);

bail_unlock:
	spin_unlock_irqrestore(&ipath_devs_lock, flags);

bail:
	return dd;
}

static inline struct ipath_devdata *__ipath_lookup(int unit)
{
	return idr_find(&unit_table, unit);
}

struct ipath_devdata *ipath_lookup(int unit)
{
	struct ipath_devdata *dd;
	unsigned long flags;

	spin_lock_irqsave(&ipath_devs_lock, flags);
	dd = __ipath_lookup(unit);
	spin_unlock_irqrestore(&ipath_devs_lock, flags);

	return dd;
}

int ipath_count_units(int *npresentp, int *nupp, u32 *maxportsp)
{
	int nunits, npresent, nup;
	struct ipath_devdata *dd;
	unsigned long flags;
	u32 maxports;

	nunits = npresent = nup = maxports = 0;

	spin_lock_irqsave(&ipath_devs_lock, flags);

	list_for_each_entry(dd, &ipath_dev_list, ipath_list) {
		nunits++;
		if ((dd->ipath_flags & IPATH_PRESENT) && dd->ipath_kregbase)
			npresent++;
		if (dd->ipath_lid &&
		    !(dd->ipath_flags & (IPATH_DISABLED | IPATH_LINKDOWN
					 | IPATH_LINKUNK)))
			nup++;
		if (dd->ipath_cfgports > maxports)
			maxports = dd->ipath_cfgports;
	}

	spin_unlock_irqrestore(&ipath_devs_lock, flags);

	if (npresentp)
		*npresentp = npresent;
	if (nupp)
		*nupp = nup;
	if (maxportsp)
		*maxportsp = maxports;

	return nunits;
}

/*
 * 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)) ipath_enable_wc(struct ipath_devdata *dd)
{
	return -EOPNOTSUPP;
}

void __attribute__((weak)) ipath_disable_wc(struct ipath_devdata *dd)
{
}

static int __devinit ipath_init_one(struct pci_dev *pdev,
				    const struct pci_device_id *ent)
{
	int ret, len, j;
	struct ipath_devdata *dd;
	unsigned long long addr;
	u32 bar0 = 0, bar1 = 0;
	u8 rev;

	dd = ipath_alloc_devdata(pdev);
	if (IS_ERR(dd)) {
		ret = PTR_ERR(dd);
		printk(KERN_ERR IPATH_DRV_NAME
		       ": Could not allocate devdata: error %d\n", -ret);
		goto bail;
	}

	ipath_cdbg(VERBOSE, "initializing unit #%u\n", dd->ipath_unit);

	read_bars(dd, pdev, &bar0, &bar1);

	ret = pci_enable_device(pdev);
	if (ret) {
		/* This can happen iff:
		 *
		 * We did a chip reset, and then failed to reprogram the
		 * BAR, or the chip reset due to an internal error.  We then
		 * unloaded the driver and reloaded it.
		 *
		 * Both reset cases set the BAR back to initial state.  For
		 * the latter case, the AER sticky error bit at offset 0x718
		 * should be set, but the Linux kernel doesn't yet know
		 * about that, it appears.  If the original BAR was retained
		 * in the kernel data structures, this may be OK.
		 */
		ipath_dev_err(dd, "enable unit %d failed: error %d\n",
			      dd->ipath_unit, -ret);
		goto bail_devdata;
	}
	addr = pci_resource_start(pdev, 0);
	len = pci_resource_len(pdev, 0);
	ipath_cdbg(VERBOSE, "regbase (0) %llx len %d irq %x, vend %x/%x "
		   "driver_data %lx\n", addr, len, pdev->irq, ent->vendor,
		   ent->device, ent->driver_data);

	read_bars(dd, pdev, &bar0, &bar1);

	if (!bar1 && !(bar0 & ~0xf)) {
		if (addr) {
			dev_info(&pdev->dev, "BAR is 0 (probable RESET), "
				 "rewriting as %llx\n", addr);
			ret = pci_write_config_dword(
				pdev, PCI_BASE_ADDRESS_0, addr);
			if (ret) {
				ipath_dev_err(dd, "rewrite of BAR0 "
					      "failed: err %d\n", -ret);
				goto bail_disable;
			}
			ret = pci_write_config_dword(
				pdev, PCI_BASE_ADDRESS_1, addr >> 32);
			if (ret) {
				ipath_dev_err(dd, "rewrite of BAR1 "
					      "failed: err %d\n", -ret);
				goto bail_disable;
			}
		} else {
			ipath_dev_err(dd, "BAR is 0 (probable RESET), "
				      "not usable until reboot\n");
			ret = -ENODEV;
			goto bail_disable;
		}
	}

	ret = pci_request_regions(pdev, IPATH_DRV_NAME);
	if (ret) {
		dev_info(&pdev->dev, "pci_request_regions unit %u fails: "
			 "err %d\n", dd->ipath_unit, -ret);
		goto bail_disable;
	}

	ret = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
	if (ret) {
		/*
		 * if the 64 bit setup fails, try 32 bit.  Some systems
		 * do not setup 64 bit maps on systems with 2GB or less
		 * memory installed.
		 */
		ret = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
		if (ret) {
			dev_info(&pdev->dev,
				"Unable to set DMA mask for unit %u: %d\n",
				dd->ipath_unit, ret);
			goto bail_regions;
		}
		else {
			ipath_dbg("No 64bit DMA mask, used 32 bit mask\n");
			ret = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
			if (ret)
				dev_info(&pdev->dev,
					"Unable to set DMA consistent mask "
					"for unit %u: %d\n",
					dd->ipath_unit, ret);

		}
	}
	else {
		ret = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
		if (ret)
			dev_info(&pdev->dev,
				"Unable to set DMA consistent mask "
				"for unit %u: %d\n",
				dd->ipath_unit, ret);
	}

	pci_set_master(pdev);

	/*
	 * Save BARs to rewrite after device reset.  Save all 64 bits of
	 * BAR, just in case.
	 */
	dd->ipath_pcibar0 = addr;
	dd->ipath_pcibar1 = addr >> 32;
	dd->ipath_deviceid = ent->device;	/* save for later use */
	dd->ipath_vendorid = ent->vendor;

	/* setup the chip-specific functions, as early as possible. */
	switch (ent->device) {
	case PCI_DEVICE_ID_INFINIPATH_HT:
		ipath_init_ht400_funcs(dd);
		break;
	case PCI_DEVICE_ID_INFINIPATH_PE800:
		ipath_init_pe800_funcs(dd);
		break;
	default:
		ipath_dev_err(dd, "Found unknown QLogic deviceid 0x%x, "
			      "failing\n", ent->device);
		return -ENODEV;
	}

	for (j = 0; j < 6; j++) {
		if (!pdev->resource[j].start)
			continue;
		ipath_cdbg(VERBOSE, "BAR %d start %llx, end %llx, len %llx\n",
			   j, (unsigned long long)pdev->resource[j].start,
			   (unsigned long long)pdev->resource[j].end,
			   (unsigned long long)pci_resource_len(pdev, j));
	}

	if (!addr) {
		ipath_dev_err(dd, "No valid address in BAR 0!\n");
		ret = -ENODEV;
		goto bail_regions;
	}

	dd->ipath_deviceid = ent->device;	/* save for later use */
	dd->ipath_vendorid = ent->vendor;

	ret = pci_read_config_byte(pdev, PCI_REVISION_ID, &rev);
	if (ret) {
		ipath_dev_err(dd, "Failed to read PCI revision ID unit "
			      "%u: err %d\n", dd->ipath_unit, -ret);
		goto bail_regions;	/* shouldn't ever happen */
	}
	dd->ipath_pcirev = rev;

	dd->ipath_kregbase = ioremap_nocache(addr, len);

	if (!dd->ipath_kregbase) {
		ipath_dbg("Unable to map io addr %llx to kvirt, failing\n",
			  addr);
		ret = -ENOMEM;
		goto bail_iounmap;
	}
	dd->ipath_kregend = (u64 __iomem *)
		((void __iomem *)dd->ipath_kregbase + len);
	dd->ipath_physaddr = addr;	/* used for io_remap, etc. */
	/* for user mmap */
	ipath_cdbg(VERBOSE, "mapped io addr %llx to kregbase %p\n",
		   addr, dd->ipath_kregbase);

	/*
	 * clear ipath_flags here instead of in ipath_init_chip as it is set
	 * by ipath_setup_htconfig.
	 */
	dd->ipath_flags = 0;

	if (dd->ipath_f_bus(dd, pdev))
		ipath_dev_err(dd, "Failed to setup config space; "
			      "continuing anyway\n");

	/*
	 * set up our interrupt handler; SA_SHIRQ probably not needed,
	 * since MSI interrupts shouldn't be shared but won't  hurt for now.
	 * check 0 irq after we return from chip-specific bus setup, since
	 * that can affect this due to setup
	 */
	if (!pdev->irq)
		ipath_dev_err(dd, "irq is 0, BIOS error?  Interrupts won't "
			      "work\n");
	else {
		ret = request_irq(pdev->irq, ipath_intr, SA_SHIRQ,
				  IPATH_DRV_NAME, dd);
		if (ret) {
			ipath_dev_err(dd, "Couldn't setup irq handler, "
				      "irq=%u: %d\n", pdev->irq, ret);
			goto bail_iounmap;
		}
	}

	ret = ipath_init_chip(dd, 0);	/* do the chip-specific init */
	if (ret)
		goto bail_iounmap;

	ret = ipath_enable_wc(dd);

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

	ipath_device_create_group(&pdev->dev, dd);
	ipathfs_add_device(dd);
	ipath_user_add(dd);
	ipath_diag_add(dd);
	ipath_layer_add(dd);

	goto bail;

bail_iounmap:
	iounmap((volatile void __iomem *) dd->ipath_kregbase);

bail_regions:
	pci_release_regions(pdev);

bail_disable:
	pci_disable_device(pdev);

bail_devdata:
	ipath_free_devdata(pdev, dd);

bail:
	return ret;
}

static void __devexit ipath_remove_one(struct pci_dev *pdev)
{
	struct ipath_devdata *dd;

	ipath_cdbg(VERBOSE, "removing, pdev=%p\n", pdev);
	if (!pdev)
		return;

	dd = pci_get_drvdata(pdev);
	ipath_layer_remove(dd);
	ipath_diag_remove(dd);
	ipath_user_remove(dd);
	ipathfs_remove_device(dd);
	ipath_device_remove_group(&pdev->dev, dd);
	ipath_cdbg(VERBOSE, "Releasing pci memory regions, dd %p, "
		   "unit %u\n", dd, (u32) dd->ipath_unit);
	if (dd->ipath_kregbase) {
		ipath_cdbg(VERBOSE, "Unmapping kregbase %p\n",
			   dd->ipath_kregbase);
		iounmap((volatile void __iomem *) dd->ipath_kregbase);
		dd->ipath_kregbase = NULL;
	}
	pci_release_regions(pdev);
	ipath_cdbg(VERBOSE, "calling pci_disable_device\n");
	pci_disable_device(pdev);

	ipath_free_devdata(pdev, dd);
}

/* general driver use */
DEFINE_MUTEX(ipath_mutex);

static DEFINE_SPINLOCK(ipath_pioavail_lock);

/**
 * ipath_disarm_piobufs - cancel a range of PIO buffers
 * @dd: the infinipath device
 * @first: the first PIO buffer to cancel
 * @cnt: the number of PIO buffers to cancel
 *
 * cancel a range of PIO buffers, used when they might be armed, but
 * not triggered.  Used at init to ensure buffer state, and also user
 * process close, in case it died while writing to a PIO buffer
 * Also after errors.
 */
void ipath_disarm_piobufs(struct ipath_devdata *dd, unsigned first,
			  unsigned cnt)
{
	unsigned i, last = first + cnt;
	u64 sendctrl, sendorig;

	ipath_cdbg(PKT, "disarm %u PIObufs first=%u\n", cnt, first);
	sendorig = dd->ipath_sendctrl | INFINIPATH_S_DISARM;
	for (i = first; i < last; i++) {
		sendctrl = sendorig |
			(i << INFINIPATH_S_DISARMPIOBUF_SHIFT);
		ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl,
				 sendctrl);
	}

	/*
	 * Write it again with current value, in case ipath_sendctrl changed
	 * while we were looping; no critical bits that would require
	 * locking.
	 *
	 * Write a 0, and then the original value, reading scratch in
	 * between.  This seems to avoid a chip timing race that causes
	 * pioavail updates to memory to stop.
	 */
	ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl,
			 0);
	sendorig = ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch);
	ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl,
			 dd->ipath_sendctrl);
}

/**
 * ipath_wait_linkstate - wait for an IB link state change to occur
 * @dd: the infinipath device
 * @state: the state to wait for
 * @msecs: the number of milliseconds to wait
 *
 * wait up to msecs milliseconds for IB link state change to occur for
 * now, take the easy polling route.  Currently used only by
 * ipath_layer_set_linkstate.  Returns 0 if state reached, otherwise
 * -ETIMEDOUT state can have multiple states set, for any of several
 * transitions.
 */
int ipath_wait_linkstate(struct ipath_devdata *dd, u32 state, int msecs)
{
	dd->ipath_sma_state_wanted = state;
	wait_event_interruptible_timeout(ipath_sma_state_wait,
					 (dd->ipath_flags & state),
					 msecs_to_jiffies(msecs));
	dd->ipath_sma_state_wanted = 0;

	if (!(dd->ipath_flags & state)) {
		u64 val;
		ipath_cdbg(SMA, "Didn't reach linkstate %s within %u ms\n",
			   /* test INIT ahead of DOWN, both can be set */
			   (state & IPATH_LINKINIT) ? "INIT" :
			   ((state & IPATH_LINKDOWN) ? "DOWN" :
			    ((state & IPATH_LINKARMED) ? "ARM" : "ACTIVE")),
			   msecs);
		val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_ibcstatus);
		ipath_cdbg(VERBOSE, "ibcc=%llx ibcstatus=%llx (%s)\n",
			   (unsigned long long) ipath_read_kreg64(
				   dd, dd->ipath_kregs->kr_ibcctrl),
			   (unsigned long long) val,
			   ipath_ibcstatus_str[val & 0xf]);
	}
	return (dd->ipath_flags & state) ? 0 : -ETIMEDOUT;
}

void ipath_decode_err(char *buf, size_t blen, ipath_err_t err)
{
	*buf = '\0';
	if (err & INFINIPATH_E_RHDRLEN)
		strlcat(buf, "rhdrlen ", blen);
	if (err & INFINIPATH_E_RBADTID)
		strlcat(buf, "rbadtid ", blen);
	if (err & INFINIPATH_E_RBADVERSION)
		strlcat(buf, "rbadversion ", blen);
	if (err & INFINIPATH_E_RHDR)
		strlcat(buf, "rhdr ", blen);
	if (err & INFINIPATH_E_RLONGPKTLEN)
		strlcat(buf, "rlongpktlen ", blen);
	if (err & INFINIPATH_E_RSHORTPKTLEN)
		strlcat(buf, "rshortpktlen ", blen);
	if (err & INFINIPATH_E_RMAXPKTLEN)
		strlcat(buf, "rmaxpktlen ", blen);
	if (err & INFINIPATH_E_RMINPKTLEN)
		strlcat(buf, "rminpktlen ", blen);
	if (err & INFINIPATH_E_RFORMATERR)
		strlcat(buf, "rformaterr ", blen);
	if (err & INFINIPATH_E_RUNSUPVL)
		strlcat(buf, "runsupvl ", blen);
	if (err & INFINIPATH_E_RUNEXPCHAR)
		strlcat(buf, "runexpchar ", blen);
	if (err & INFINIPATH_E_RIBFLOW)
		strlcat(buf, "ribflow ", blen);
	if (err & INFINIPATH_E_REBP)
		strlcat(buf, "EBP ", blen);
	if (err & INFINIPATH_E_SUNDERRUN)
		strlcat(buf, "sunderrun ", blen);
	if (err & INFINIPATH_E_SPIOARMLAUNCH)
		strlcat(buf, "spioarmlaunch ", blen);
	if (err & INFINIPATH_E_SUNEXPERRPKTNUM)
		strlcat(buf, "sunexperrpktnum ", blen);
	if (err & INFINIPATH_E_SDROPPEDDATAPKT)
		strlcat(buf, "sdroppeddatapkt ", blen);
	if (err & INFINIPATH_E_SDROPPEDSMPPKT)
		strlcat(buf, "sdroppedsmppkt ", blen);
	if (err & INFINIPATH_E_SMAXPKTLEN)
		strlcat(buf, "smaxpktlen ", blen);
	if (err & INFINIPATH_E_SMINPKTLEN)
		strlcat(buf, "sminpktlen ", blen);
	if (err & INFINIPATH_E_SUNSUPVL)
		strlcat(buf, "sunsupVL ", blen);
	if (err & INFINIPATH_E_SPKTLEN)
		strlcat(buf, "spktlen ", blen);
	if (err & INFINIPATH_E_INVALIDADDR)
		strlcat(buf, "invalidaddr ", blen);
	if (err & INFINIPATH_E_RICRC)
		strlcat(buf, "CRC ", blen);
	if (err & INFINIPATH_E_RVCRC)
		strlcat(buf, "VCRC ", blen);
	if (err & INFINIPATH_E_RRCVEGRFULL)
		strlcat(buf, "rcvegrfull ", blen);
	if (err & INFINIPATH_E_RRCVHDRFULL)
		strlcat(buf, "rcvhdrfull ", blen);
	if (err & INFINIPATH_E_IBSTATUSCHANGED)
		strlcat(buf, "ibcstatuschg ", blen);
	if (err & INFINIPATH_E_RIBLOSTLINK)
		strlcat(buf, "riblostlink ", blen);
	if (err & INFINIPATH_E_HARDWARE)
		strlcat(buf, "hardware ", blen);
	if (err & INFINIPATH_E_RESET)
		strlcat(buf, "reset ", blen);
}

/**
 * get_rhf_errstring - decode RHF errors
 * @err: the err number
 * @msg: the output buffer
 * @len: the length of the output buffer
 *
 * only used one place now, may want more later
 */
static void get_rhf_errstring(u32 err, char *msg, size_t len)
{
	/* if no errors, and so don't need to check what's first */
	*msg = '\0';

	if (err & INFINIPATH_RHF_H_ICRCERR)
		strlcat(msg, "icrcerr ", len);
	if (err & INFINIPATH_RHF_H_VCRCERR)
		strlcat(msg, "vcrcerr ", len);
	if (err & INFINIPATH_RHF_H_PARITYERR)
		strlcat(msg, "parityerr ", len);
	if (err & INFINIPATH_RHF_H_LENERR)
		strlcat(msg, "lenerr ", len);
	if (err & INFINIPATH_RHF_H_MTUERR)
		strlcat(msg, "mtuerr ", len);
	if (err & INFINIPATH_RHF_H_IHDRERR)
		/* infinipath hdr checksum error */
		strlcat(msg, "ipathhdrerr ", len);
	if (err & INFINIPATH_RHF_H_TIDERR)
		strlcat(msg, "tiderr ", len);
	if (err & INFINIPATH_RHF_H_MKERR)
		/* bad port, offset, etc. */
		strlcat(msg, "invalid ipathhdr ", len);
	if (err & INFINIPATH_RHF_H_IBERR)
		strlcat(msg, "iberr ", len);
	if (err & INFINIPATH_RHF_L_SWA)
		strlcat(msg, "swA ", len);
	if (err & INFINIPATH_RHF_L_SWB)
		strlcat(msg, "swB ", len);
}

/**
 * ipath_get_egrbuf - get an eager buffer
 * @dd: the infinipath device
 * @bufnum: the eager buffer to get
 * @err: unused
 *
 * must only be called if ipath_pd[port] is known to be allocated
 */
static inline void *ipath_get_egrbuf(struct ipath_devdata *dd, u32 bufnum,
				     int err)
{
	return dd->ipath_port0_skbs ?
		(void *)dd->ipath_port0_skbs[bufnum]->data : NULL;
}

/**
 * ipath_alloc_skb - allocate an skb and buffer with possible constraints
 * @dd: the infinipath device
 * @gfp_mask: the sk_buff SFP mask
 */
struct sk_buff *ipath_alloc_skb(struct ipath_devdata *dd,
				gfp_t gfp_mask)
{
	struct sk_buff *skb;
	u32 len;

	/*
	 * Only fully supported way to handle this is to allocate lots
	 * extra, align as needed, and then do skb_reserve().  That wastes
	 * a lot of memory...  I'll have to hack this into infinipath_copy
	 * also.
	 */

	/*
	 * We need 4 extra bytes for unaligned transfer copying
	 */
	if (dd->ipath_flags & IPATH_4BYTE_TID) {
		/* we need a 4KB multiple alignment, and there is no way
		 * to do it except to allocate extra and then skb_reserve
		 * enough to bring it up to the right alignment.
		 */
		len = dd->ipath_ibmaxlen + 4 + (1 << 11) - 1;
	}
	else
		len = dd->ipath_ibmaxlen + 4;
	skb = __dev_alloc_skb(len, gfp_mask);
	if (!skb) {
		ipath_dev_err(dd, "Failed to allocate skbuff, length %u\n",
			      len);
		goto bail;
	}
	if (dd->ipath_flags & IPATH_4BYTE_TID) {
		u32 una = ((1 << 11) - 1) & (unsigned long)(skb->data + 4);
		if (una)
			skb_reserve(skb, 4 + (1 << 11) - una);
		else
			skb_reserve(skb, 4);
	} else
		skb_reserve(skb, 4);

bail:
	return skb;
}

/**
 * ipath_rcv_layer - receive a packet for the layered (ethernet) driver
 * @dd: the infinipath device
 * @etail: the sk_buff number
 * @tlen: the total packet length
 * @hdr: the ethernet header
 *
 * Separate routine for better overall optimization
 */
static void ipath_rcv_layer(struct ipath_devdata *dd, u32 etail,
			    u32 tlen, struct ether_header *hdr)
{
	u32 elen;
	u8 pad, *bthbytes;
	struct sk_buff *skb, *nskb;

	if (dd->ipath_port0_skbs && hdr->sub_opcode == OPCODE_ENCAP) {
		/*
		 * Allocate a new sk_buff to replace the one we give
		 * to the network stack.
		 */
		nskb = ipath_alloc_skb(dd, GFP_ATOMIC);
		if (!nskb) {
			/* count OK packets that we drop */
			ipath_stats.sps_krdrops++;
			return;
		}

		bthbytes = (u8 *) hdr->bth;
		pad = (bthbytes[1] >> 4) & 3;
		/* +CRC32 */
		elen = tlen - (sizeof(*hdr) + pad + sizeof(u32));

		skb = dd->ipath_port0_skbs[etail];
		dd->ipath_port0_skbs[etail] = nskb;
		skb_put(skb, elen);

		dd->ipath_f_put_tid(dd, etail + (u64 __iomem *)
				    ((char __iomem *) dd->ipath_kregbase
				     + dd->ipath_rcvegrbase), 0,
				    virt_to_phys(nskb->data));

		__ipath_layer_rcv(dd, hdr, skb);

		/* another ether packet received */
		ipath_stats.sps_ether_rpkts++;
	}
	else if (hdr->sub_opcode == OPCODE_LID_ARP)
		__ipath_layer_rcv_lid(dd, hdr);
}

/*
 * ipath_kreceive - receive a packet
 * @dd: the infinipath device
 *
 * called from interrupt handler for errors or receive interrupt
 */
void ipath_kreceive(struct ipath_devdata *dd)
{
	u64 *rc;
	void *ebuf;
	const u32 rsize = dd->ipath_rcvhdrentsize;	/* words */
	const u32 maxcnt = dd->ipath_rcvhdrcnt * rsize;	/* words */
	u32 etail = -1, l, hdrqtail;
	struct ips_message_header *hdr;
	u32 eflags, i, etype, tlen, pkttot = 0;
	static u64 totcalls;	/* stats, may eventually remove */
	char emsg[128];

	if (!dd->ipath_hdrqtailptr) {
		ipath_dev_err(dd,
			      "hdrqtailptr not set, can't do receives\n");
		goto bail;
	}

	/* There is already a thread processing this queue. */
	if (test_and_set_bit(0, &dd->ipath_rcv_pending))
		goto bail;

	if (dd->ipath_port0head ==
	    (u32)le64_to_cpu(*dd->ipath_hdrqtailptr))
		goto done;

	/* read only once at start for performance */
	hdrqtail = (u32)le64_to_cpu(*dd->ipath_hdrqtailptr);

	for (i = 0, l = dd->ipath_port0head; l != hdrqtail; i++) {
		u32 qp;
		u8 *bthbytes;

		rc = (u64 *) (dd->ipath_pd[0]->port_rcvhdrq + (l << 2));
		hdr = (struct ips_message_header *)&rc[1];
		/*
		 * could make a network order version of IPATH_KD_QP, and
		 * do the obvious shift before masking to speed this up.
		 */
		qp = ntohl(hdr->bth[1]) & 0xffffff;
		bthbytes = (u8 *) hdr->bth;

		eflags = ips_get_hdr_err_flags((__le32 *) rc);
		etype = ips_get_rcv_type((__le32 *) rc);
		/* total length */
		tlen = ips_get_length_in_bytes((__le32 *) rc);
		ebuf = NULL;
		if (etype != RCVHQ_RCV_TYPE_EXPECTED) {
			/*
			 * it turns out that the chips uses an eager buffer
			 * for all non-expected packets, whether it "needs"
			 * one or not.  So always get the index, but don't
			 * set ebuf (so we try to copy data) unless the
			 * length requires it.
			 */
			etail = ips_get_index((__le32 *) rc);
			if (tlen > sizeof(*hdr) ||
			    etype == RCVHQ_RCV_TYPE_NON_KD)
				ebuf = ipath_get_egrbuf(dd, etail, 0);
		}

		/*
		 * both tiderr and ipathhdrerr are set for all plain IB
		 * packets; only ipathhdrerr should be set.
		 */

		if (etype != RCVHQ_RCV_TYPE_NON_KD && etype !=
		    RCVHQ_RCV_TYPE_ERROR && ips_get_ipath_ver(
			    hdr->iph.ver_port_tid_offset) !=
		    IPS_PROTO_VERSION) {
			ipath_cdbg(PKT, "Bad InfiniPath protocol version "
				   "%x\n", etype);
		}

		if (eflags & ~(INFINIPATH_RHF_H_TIDERR |
			       INFINIPATH_RHF_H_IHDRERR)) {
			get_rhf_errstring(eflags, emsg, sizeof emsg);
			ipath_cdbg(PKT, "RHFerrs %x hdrqtail=%x typ=%u "
				   "tlen=%x opcode=%x egridx=%x: %s\n",
				   eflags, l, etype, tlen, bthbytes[0],
				   ips_get_index((__le32 *) rc), emsg);
		} else if (etype == RCVHQ_RCV_TYPE_NON_KD) {
				int ret = __ipath_verbs_rcv(dd, rc + 1,
							    ebuf, tlen);
				if (ret == -ENODEV)
					ipath_cdbg(VERBOSE,
						   "received IB packet, "
						   "not SMA (QP=%x)\n", qp);
		} else if (etype == RCVHQ_RCV_TYPE_EAGER) {
			if (qp == IPATH_KD_QP &&
			    bthbytes[0] == ipath_layer_rcv_opcode &&
			    ebuf)
				ipath_rcv_layer(dd, etail, tlen,
						(struct ether_header *)hdr);
			else
				ipath_cdbg(PKT, "typ %x, opcode %x (eager, "
					   "qp=%x), len %x; ignored\n",
					   etype, bthbytes[0], qp, tlen);
		}
		else if (etype == RCVHQ_RCV_TYPE_EXPECTED)
			ipath_dbg("Bug: Expected TID, opcode %x; ignored\n",
				  be32_to_cpu(hdr->bth[0]) & 0xff);
		else if (eflags & (INFINIPATH_RHF_H_TIDERR |
				   INFINIPATH_RHF_H_IHDRERR)) {
			/*
			 * This is a type 3 packet, only the LRH is in the
			 * rcvhdrq, the rest of the header is in the eager
			 * buffer.
			 */
			u8 opcode;
			if (ebuf) {
				bthbytes = (u8 *) ebuf;
				opcode = *bthbytes;
			}
			else
				opcode = 0;
			get_rhf_errstring(eflags, emsg, sizeof emsg);
			ipath_dbg("Err %x (%s), opcode %x, egrbuf %x, "
				  "len %x\n", eflags, emsg, opcode, etail,
				  tlen);
		} else {
			/*
			 * error packet, type of error	unknown.
			 * Probably type 3, but we don't know, so don't
			 * even try to print the opcode, etc.
			 */
			ipath_dbg("Error Pkt, but no eflags! egrbuf %x, "
				  "len %x\nhdrq@%lx;hdrq+%x rhf: %llx; "
				  "hdr %llx %llx %llx %llx %llx\n",
				  etail, tlen, (unsigned long) rc, l,
				  (unsigned long long) rc[0],
				  (unsigned long long) rc[1],
				  (unsigned long long) rc[2],
				  (unsigned long long) rc[3],
				  (unsigned long long) rc[4],
				  (unsigned long long) rc[5]);
		}
		l += rsize;
		if (l >= maxcnt)
			l = 0;
		/*
		 * update for each packet, to help prevent overflows if we
		 * have lots of packets.
		 */
		(void)ipath_write_ureg(dd, ur_rcvhdrhead,
				       dd->ipath_rhdrhead_intr_off | l, 0);
		if (etype != RCVHQ_RCV_TYPE_EXPECTED)
			(void)ipath_write_ureg(dd, ur_rcvegrindexhead,
					       etail, 0);
	}

	pkttot += i;

	dd->ipath_port0head = l;

	if (pkttot > ipath_stats.sps_maxpkts_call)
		ipath_stats.sps_maxpkts_call = pkttot;
	ipath_stats.sps_port0pkts += pkttot;
	ipath_stats.sps_avgpkts_call =
		ipath_stats.sps_port0pkts / ++totcalls;

done:
	clear_bit(0, &dd->ipath_rcv_pending);
	smp_mb__after_clear_bit();

bail:;
}

/**
 * ipath_update_pio_bufs - update shadow copy of the PIO availability map
 * @dd: the infinipath device
 *
 * called whenever our local copy indicates we have run out of send buffers
 * NOTE: This can be called from interrupt context by some code
 * and from non-interrupt context by ipath_getpiobuf().
 */

static void ipath_update_pio_bufs(struct ipath_devdata *dd)
{
	unsigned long flags;
	int i;
	const unsigned piobregs = (unsigned)dd->ipath_pioavregs;

	/* If the generation (check) bits have changed, then we update the
	 * busy bit for the corresponding PIO buffer.  This algorithm will
	 * modify positions to the value they already have in some cases
	 * (i.e., no change), but it's faster than changing only the bits
	 * that have changed.
	 *
	 * We would like to do this atomicly, to avoid spinlocks in the
	 * critical send path, but that's not really possible, given the
	 * type of changes, and that this routine could be called on
	 * multiple cpu's simultaneously, so we lock in this routine only,
	 * to avoid conflicting updates; all we change is the shadow, and
	 * it's a single 64 bit memory location, so by definition the update
	 * is atomic in terms of what other cpu's can see in testing the
	 * bits.  The spin_lock overhead isn't too bad, since it only
	 * happens when all buffers are in use, so only cpu overhead, not
	 * latency or bandwidth is affected.
	 */
#define _IPATH_ALL_CHECKBITS 0x5555555555555555ULL
	if (!dd->ipath_pioavailregs_dma) {
		ipath_dbg("Update shadow pioavail, but regs_dma NULL!\n");
		return;
	}
	if (ipath_debug & __IPATH_VERBDBG) {
		/* only if packet debug and verbose */
		volatile __le64 *dma = dd->ipath_pioavailregs_dma;
		unsigned long *shadow = dd->ipath_pioavailshadow;

		ipath_cdbg(PKT, "Refill avail, dma0=%llx shad0=%lx, "
			   "d1=%llx s1=%lx, d2=%llx s2=%lx, d3=%llx "
			   "s3=%lx\n",
			   (unsigned long long) le64_to_cpu(dma[0]),
			   shadow[0],
			   (unsigned long long) le64_to_cpu(dma[1]),
			   shadow[1],
			   (unsigned long long) le64_to_cpu(dma[2]),
			   shadow[2],
			   (unsigned long long) le64_to_cpu(dma[3]),
			   shadow[3]);
		if (piobregs > 4)
			ipath_cdbg(
				PKT, "2nd group, dma4=%llx shad4=%lx, "
				"d5=%llx s5=%lx, d6=%llx s6=%lx, "
				"d7=%llx s7=%lx\n",
				(unsigned long long) le64_to_cpu(dma[4]),
				shadow[4],
				(unsigned long long) le64_to_cpu(dma[5]),
				shadow[5],
				(unsigned long long) le64_to_cpu(dma[6]),
				shadow[6],
				(unsigned long long) le64_to_cpu(dma[7]),
				shadow[7]);
	}
	spin_lock_irqsave(&ipath_pioavail_lock, flags);
	for (i = 0; i < piobregs; i++) {
		u64 pchbusy, pchg, piov, pnew;
		/*
		 * Chip Errata: bug 6641; even and odd qwords>3 are swapped
		 */
		if (i > 3) {
			if (i & 1)
				piov = le64_to_cpu(
					dd->ipath_pioavailregs_dma[i - 1]);
			else
				piov = le64_to_cpu(
					dd->ipath_pioavailregs_dma[i + 1]);
		} else
			piov = le64_to_cpu(dd->ipath_pioavailregs_dma[i]);
		pchg = _IPATH_ALL_CHECKBITS &
			~(dd->ipath_pioavailshadow[i] ^ piov);
		pchbusy = pchg << INFINIPATH_SENDPIOAVAIL_BUSY_SHIFT;
		if (pchg && (pchbusy & dd->ipath_pioavailshadow[i])) {
			pnew = dd->ipath_pioavailshadow[i] & ~pchbusy;
			pnew |= piov & pchbusy;
			dd->ipath_pioavailshadow[i] = pnew;
		}
	}
	spin_unlock_irqrestore(&ipath_pioavail_lock, flags);
}

/**
 * ipath_setrcvhdrsize - set the receive header size
 * @dd: the infinipath device
 * @rhdrsize: the receive header size
 *
 * called from user init code, and also layered driver init
 */
int ipath_setrcvhdrsize(struct ipath_devdata *dd, unsigned rhdrsize)
{
	int ret = 0;

	if (dd->ipath_flags & IPATH_RCVHDRSZ_SET) {
		if (dd->ipath_rcvhdrsize != rhdrsize) {
			dev_info(&dd->pcidev->dev,
				 "Error: can't set protocol header "
				 "size %u, already %u\n",
				 rhdrsize, dd->ipath_rcvhdrsize);
			ret = -EAGAIN;
		} else
			ipath_cdbg(VERBOSE, "Reuse same protocol header "
				   "size %u\n", dd->ipath_rcvhdrsize);
	} else if (rhdrsize > (dd->ipath_rcvhdrentsize -
			       (sizeof(u64) / sizeof(u32)))) {
		ipath_dbg("Error: can't set protocol header size %u "
			  "(> max %u)\n", rhdrsize,
			  dd->ipath_rcvhdrentsize -
			  (u32) (sizeof(u64) / sizeof(u32)));
		ret = -EOVERFLOW;
	} else {
		dd->ipath_flags |= IPATH_RCVHDRSZ_SET;
		dd->ipath_rcvhdrsize = rhdrsize;
		ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvhdrsize,
				 dd->ipath_rcvhdrsize);
		ipath_cdbg(VERBOSE, "Set protocol header size to %u\n",
			   dd->ipath_rcvhdrsize);
	}
	return ret;
}

/**
 * ipath_getpiobuf - find an available pio buffer
 * @dd: the infinipath device
 * @pbufnum: the buffer number is placed here
 *
 * do appropriate marking as busy, etc.
 * returns buffer number if one found (>=0), negative number is error.
 * Used by ipath_sma_send_pkt and ipath_layer_send
 */
u32 __iomem *ipath_getpiobuf(struct ipath_devdata *dd, u32 * pbufnum)
{
	int i, j, starti, updated = 0;
	unsigned piobcnt, iter;
	unsigned long flags;
	unsigned long *shadow = dd->ipath_pioavailshadow;
	u32 __iomem *buf;

	piobcnt = (unsigned)(dd->ipath_piobcnt2k
			     + dd->ipath_piobcnt4k);
	starti = dd->ipath_lastport_piobuf;
	iter = piobcnt - starti;
	if (dd->ipath_upd_pio_shadow) {
		/*
		 * Minor optimization.  If we had no buffers on last call,
		 * start out by doing the update; continue and do scan even
		 * if no buffers were updated, to be paranoid
		 */
		ipath_update_pio_bufs(dd);
		/* we scanned here, don't do it at end of scan */
		updated = 1;
		i = starti;
	} else
		i = dd->ipath_lastpioindex;

rescan:
	/*
	 * while test_and_set_bit() is atomic, we do that and then the
	 * change_bit(), and the pair is not.  See if this is the cause
	 * of the remaining armlaunch errors.
	 */
	spin_lock_irqsave(&ipath_pioavail_lock, flags);
	for (j = 0; j < iter; j++, i++) {
		if (i >= piobcnt)
			i = starti;
		/*
		 * To avoid bus lock overhead, we first find a candidate
		 * buffer, then do the test and set, and continue if that
		 * fails.
		 */
		if (test_bit((2 * i) + 1, shadow) ||
		    test_and_set_bit((2 * i) + 1, shadow))
			continue;
		/* flip generation bit */
		change_bit(2 * i, shadow);
		break;
	}
	spin_unlock_irqrestore(&ipath_pioavail_lock, flags);

	if (j == iter) {
		volatile __le64 *dma = dd->ipath_pioavailregs_dma;

		/*
		 * first time through; shadow exhausted, but may be real
		 * buffers available, so go see; if any updated, rescan
		 * (once)
		 */
		if (!updated) {
			ipath_update_pio_bufs(dd);
			updated = 1;
			i = starti;
			goto rescan;
		}
		dd->ipath_upd_pio_shadow = 1;
		/*
		 * not atomic, but if we lose one once in a while, that's OK
		 */
		ipath_stats.sps_nopiobufs++;
		if (!(++dd->ipath_consec_nopiobuf % 100000)) {
			ipath_dbg(
				"%u pio sends with no bufavail; dmacopy: "
				"%llx %llx %llx %llx; shadow:  "
				"%lx %lx %lx %lx\n",
				dd->ipath_consec_nopiobuf,
				(unsigned long long) le64_to_cpu(dma[0]),
				(unsigned long long) le64_to_cpu(dma[1]),
				(unsigned long long) le64_to_cpu(dma[2]),
				(unsigned long long) le64_to_cpu(dma[3]),
				shadow[0], shadow[1], shadow[2],
				shadow[3]);
			/*
			 * 4 buffers per byte, 4 registers above, cover rest
			 * below
			 */
			if ((dd->ipath_piobcnt2k + dd->ipath_piobcnt4k) >
			    (sizeof(shadow[0]) * 4 * 4))
				ipath_dbg("2nd group: dmacopy: %llx %llx "
					  "%llx %llx; shadow: %lx %lx "
					  "%lx %lx\n",
					  (unsigned long long)
					  le64_to_cpu(dma[4]),
					  (unsigned long long)
					  le64_to_cpu(dma[5]),
					  (unsigned long long)
					  le64_to_cpu(dma[6]),
					  (unsigned long long)
					  le64_to_cpu(dma[7]),
					  shadow[4], shadow[5],
					  shadow[6], shadow[7]);
		}
		buf = NULL;
		goto bail;
	}

	if (updated)
		/*
		 * ran out of bufs, now some (at least this one we just
		 * got) are now available, so tell the layered driver.
		 */
		__ipath_layer_intr(dd, IPATH_LAYER_INT_SEND_CONTINUE);

	/*
	 * set next starting place.  Since it's just an optimization,
	 * it doesn't matter who wins on this, so no locking
	 */
	dd->ipath_lastpioindex = i + 1;
	if (dd->ipath_upd_pio_shadow)
		dd->ipath_upd_pio_shadow = 0;
	if (dd->ipath_consec_nopiobuf)
		dd->ipath_consec_nopiobuf = 0;
	if (i < dd->ipath_piobcnt2k)
		buf = (u32 __iomem *) (dd->ipath_pio2kbase +
				       i * dd->ipath_palign);
	else
		buf = (u32 __iomem *)
			(dd->ipath_pio4kbase +
			 (i - dd->ipath_piobcnt2k) * dd->ipath_4kalign);
	ipath_cdbg(VERBOSE, "Return piobuf%u %uk @ %p\n",
		   i, (i < dd->ipath_piobcnt2k) ? 2 : 4, buf);
	if (pbufnum)
		*pbufnum = i;

bail:
	return buf;
}

/**
 * ipath_create_rcvhdrq - create a receive header queue
 * @dd: the infinipath device
 * @pd: the port 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 ipath_create_rcvhdrq(struct ipath_devdata *dd,
			 struct ipath_portdata *pd)
{
	int ret = 0;

	if (!pd->port_rcvhdrq) {
		dma_addr_t phys_hdrqtail;
		gfp_t gfp_flags = GFP_USER | __GFP_COMP;
		int amt = ALIGN(dd->ipath_rcvhdrcnt * dd->ipath_rcvhdrentsize *
				sizeof(u32), PAGE_SIZE);

		pd->port_rcvhdrq = dma_alloc_coherent(
			&dd->pcidev->dev, amt, &pd->port_rcvhdrq_phys,
			gfp_flags);

		if (!pd->port_rcvhdrq) {
			ipath_dev_err(dd, "attempt to allocate %d bytes "
				      "for port %u rcvhdrq failed\n",
				      amt, pd->port_port);
			ret = -ENOMEM;
			goto bail;
		}
		pd->port_rcvhdrtail_kvaddr = dma_alloc_coherent(
			&dd->pcidev->dev, PAGE_SIZE, &phys_hdrqtail, GFP_KERNEL);
		if (!pd->port_rcvhdrtail_kvaddr) {
			ipath_dev_err(dd, "attempt to allocate 1 page "
				      "for port %u rcvhdrqtailaddr failed\n",
				      pd->port_port);
			ret = -ENOMEM;
			goto bail;
		}
		pd->port_rcvhdrqtailaddr_phys = phys_hdrqtail;

		pd->port_rcvhdrq_size = amt;

		ipath_cdbg(VERBOSE, "%d pages at %p (phys %lx) size=%lu "
			   "for port %u rcvhdr Q\n",
			   amt >> PAGE_SHIFT, pd->port_rcvhdrq,
			   (unsigned long) pd->port_rcvhdrq_phys,
			   (unsigned long) pd->port_rcvhdrq_size,
			   pd->port_port);

		ipath_cdbg(VERBOSE, "port %d hdrtailaddr, %llx physical\n",
			   pd->port_port,
			   (unsigned long long) phys_hdrqtail);
	}
	else
		ipath_cdbg(VERBOSE, "reuse port %d rcvhdrq @%p %llx phys; "
			   "hdrtailaddr@%p %llx physical\n",
			   pd->port_port, pd->port_rcvhdrq,
			   pd->port_rcvhdrq_phys, pd->port_rcvhdrtail_kvaddr,
			   (unsigned long long)pd->port_rcvhdrqtailaddr_phys);

	/* clear for security and sanity on each use */
	memset(pd->port_rcvhdrq, 0, pd->port_rcvhdrq_size);
	memset((void *)pd->port_rcvhdrtail_kvaddr, 0, PAGE_SIZE);

	/*
	 * tell chip each time we init it, even if we are re-using previous
	 * memory (we zero the register at process close)
	 */
	ipath_write_kreg_port(dd, dd->ipath_kregs->kr_rcvhdrtailaddr,
			      pd->port_port, pd->port_rcvhdrqtailaddr_phys);
	ipath_write_kreg_port(dd, dd->ipath_kregs->kr_rcvhdraddr,
			      pd->port_port, pd->port_rcvhdrq_phys);

	ret = 0;
bail:
	return ret;
}

int ipath_waitfor_complete(struct ipath_devdata *dd, ipath_kreg reg_id,
			   u64 bits_to_wait_for, u64 * valp)
{
	unsigned long timeout;
	u64 lastval, val;
	int ret;

	lastval = ipath_read_kreg64(dd, reg_id);
	/* wait a ridiculously long time */
	timeout = jiffies + msecs_to_jiffies(5);
	do {
		val = ipath_read_kreg64(dd, reg_id);
		/* set so they have something, even on failures. */
		*valp = val;
		if ((val & bits_to_wait_for) == bits_to_wait_for) {
			ret = 0;
			break;
		}
		if (val != lastval)
			ipath_cdbg(VERBOSE, "Changed from %llx to %llx, "
				   "waiting for %llx bits\n",
				   (unsigned long long) lastval,
				   (unsigned long long) val,
				   (unsigned long long) bits_to_wait_for);
		cond_resched();
		if (time_after(jiffies, timeout)) {
			ipath_dbg("Didn't get bits %llx in register 0x%x, "
				  "got %llx\n",
				  (unsigned long long) bits_to_wait_for,
				  reg_id, (unsigned long long) *valp);
			ret = -ENODEV;
			break;
		}
	} while (1);

	return ret;
}

/**
 * ipath_waitfor_mdio_cmdready - wait for last command to complete
 * @dd: the infinipath device
 *
 * Like ipath_waitfor_complete(), but we wait for the CMDVALID bit to go
 * away indicating the last command has completed.  It doesn't return data
 */
int ipath_waitfor_mdio_cmdready(struct ipath_devdata *dd)
{
	unsigned long timeout;
	u64 val;
	int ret;

	/* wait a ridiculously long time */
	timeout = jiffies + msecs_to_jiffies(5);
	do {
		val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_mdio);
		if (!(val & IPATH_MDIO_CMDVALID)) {
			ret = 0;
			break;
		}
		cond_resched();
		if (time_after(jiffies, timeout)) {
			ipath_dbg("CMDVALID stuck in mdio reg? (%llx)\n",
				  (unsigned long long) val);
			ret = -ENODEV;
			break;
		}
	} while (1);

	return ret;
}

void ipath_set_ib_lstate(struct ipath_devdata *dd, int which)
{
	static const char *what[4] = {
		[0] = "DOWN",
		[INFINIPATH_IBCC_LINKCMD_INIT] = "INIT",
		[INFINIPATH_IBCC_LINKCMD_ARMED] = "ARMED",
		[INFINIPATH_IBCC_LINKCMD_ACTIVE] = "ACTIVE"
	};
	int linkcmd = (which >> INFINIPATH_IBCC_LINKCMD_SHIFT) &
			INFINIPATH_IBCC_LINKCMD_MASK;

	ipath_cdbg(SMA, "Trying to move unit %u to %s, current ltstate "
		   "is %s\n", dd->ipath_unit,
		   what[linkcmd],
		   ipath_ibcstatus_str[
			   (ipath_read_kreg64
			    (dd, dd->ipath_kregs->kr_ibcstatus) >>
			    INFINIPATH_IBCS_LINKTRAININGSTATE_SHIFT) &
			   INFINIPATH_IBCS_LINKTRAININGSTATE_MASK]);
	/* flush all queued sends when going to DOWN or INIT, to be sure that
	 * they don't block SMA and other MAD packets */
	if (!linkcmd || linkcmd == INFINIPATH_IBCC_LINKCMD_INIT) {
		ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl,
				 INFINIPATH_S_ABORT);
		ipath_disarm_piobufs(dd, dd->ipath_lastport_piobuf,
		                    (unsigned)(dd->ipath_piobcnt2k +
				    dd->ipath_piobcnt4k) -
				    dd->ipath_lastport_piobuf);
	}

	ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl,
			 dd->ipath_ibcctrl | which);
}

/**
 * ipath_read_kreg64_port - read a device's per-port 64-bit kernel register
 * @dd: the infinipath device
 * @regno: the register number to read
 * @port: the port containing the register
 *
 * Registers that vary with the chip implementation constants (port)
 * use this routine.
 */
u64 ipath_read_kreg64_port(const struct ipath_devdata *dd, ipath_kreg regno,
			   unsigned port)
{
	u16 where;

	if (port < dd->ipath_portcnt &&
	    (regno == dd->ipath_kregs->kr_rcvhdraddr ||
	     regno == dd->ipath_kregs->kr_rcvhdrtailaddr))
		where = regno + port;
	else
		where = -1;

	return ipath_read_kreg64(dd, where);
}

/**
 * ipath_write_kreg_port - write a device's per-port 64-bit kernel register
 * @dd: the infinipath device
 * @regno: the register number to write
 * @port: the port containing the register
 * @value: the value to write
 *
 * Registers that vary with the chip implementation constants (port)
 * use this routine.
 */
void ipath_write_kreg_port(const struct ipath_devdata *dd, ipath_kreg regno,
			  unsigned port, u64 value)
{
	u16 where;

	if (port < dd->ipath_portcnt &&
	    (regno == dd->ipath_kregs->kr_rcvhdraddr ||
	     regno == dd->ipath_kregs->kr_rcvhdrtailaddr))
		where = regno + port;
	else
		where = -1;

	ipath_write_kreg(dd, where, value);
}

/**
 * ipath_shutdown_device - shut down a device
 * @dd: the infinipath 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 ipath_init_chip(dd,1)
 */
void ipath_shutdown_device(struct ipath_devdata *dd)
{
	u64 val;

	ipath_dbg("Shutting down the device\n");

	dd->ipath_flags |= IPATH_LINKUNK;
	dd->ipath_flags &= ~(IPATH_INITTED | IPATH_LINKDOWN |
			     IPATH_LINKINIT | IPATH_LINKARMED |
			     IPATH_LINKACTIVE);
	*dd->ipath_statusp &= ~(IPATH_STATUS_IB_CONF |
				IPATH_STATUS_IB_READY);

	/* mask interrupts, but not errors */
	ipath_write_kreg(dd, dd->ipath_kregs->kr_intmask, 0ULL);

	dd->ipath_rcvctrl = 0;
	ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvctrl,
			 dd->ipath_rcvctrl);

	/*
	 * gracefully stop all sends allowing any in progress to trickle out
	 * first.
	 */
	ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl, 0ULL);
	/* flush it */
	val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch);
	/*
	 * enough for anything that's going to trickle out to have actually
	 * done so.
	 */
	udelay(5);

	/*
	 * abort any armed or launched PIO buffers that didn't go. (self
	 * clearing).  Will cause any packet currently being transmitted to
	 * go out with an EBP, and may also cause a short packet error on
	 * the receiver.
	 */
	ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl,
			 INFINIPATH_S_ABORT);

	ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKINITCMD_DISABLE <<
			    INFINIPATH_IBCC_LINKINITCMD_SHIFT);

	/*
	 * we are shutting down, so tell the layered driver.  We don't do
	 * this on just a link state change, much like ethernet, a cable
	 * unplug, etc. doesn't change driver state
	 */
	ipath_layer_intr(dd, IPATH_LAYER_INT_IF_DOWN);

	/* disable IBC */
	dd->ipath_control &= ~INFINIPATH_C_LINKENABLE;
	ipath_write_kreg(dd, dd->ipath_kregs->kr_control,
			 dd->ipath_control | INFINIPATH_C_FREEZEMODE);

	/*
	 * clear SerdesEnable and turn the leds off; do this here because
	 * we are unloading, so don't count on interrupts to move along
	 * Turn the LEDs off explictly for the same reason.
	 */
	dd->ipath_f_quiet_serdes(dd);
	dd->ipath_f_setextled(dd, 0, 0);

	if (dd->ipath_stats_timer_active) {
		del_timer_sync(&dd->ipath_stats_timer);
		dd->ipath_stats_timer_active = 0;
	}

	/*
	 * clear all interrupts and errors, so that the next time the driver
	 * is loaded or device is enabled, we know that whatever is set
	 * happened while we were unloaded
	 */
	ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrclear,
			 ~0ULL & ~INFINIPATH_HWE_MEMBISTFAILED);
	ipath_write_kreg(dd, dd->ipath_kregs->kr_errorclear, -1LL);
	ipath_write_kreg(dd, dd->ipath_kregs->kr_intclear, -1LL);
}

/**
 * ipath_free_pddata - free a port's allocated data
 * @dd: the infinipath device
 * @pd: the portdata structure
 *
 * free up any allocated data for a port
 * This should not touch anything that would affect a simultaneous
 * re-allocation of port data, because it is called after ipath_mutex
 * is released (and can be called from reinit as well).
 * It should never change any chip state, or global driver state.
 * (The only exception to global state is freeing the port0 port0_skbs.)
 */
void ipath_free_pddata(struct ipath_devdata *dd, struct ipath_portdata *pd)
{
	if (!pd)
		return;

	if (pd->port_rcvhdrq) {
		ipath_cdbg(VERBOSE, "free closed port %d rcvhdrq @ %p "
			   "(size=%lu)\n", pd->port_port, pd->port_rcvhdrq,
			   (unsigned long) pd->port_rcvhdrq_size);
		dma_free_coherent(&dd->pcidev->dev, pd->port_rcvhdrq_size,
				  pd->port_rcvhdrq, pd->port_rcvhdrq_phys);
		pd->port_rcvhdrq = NULL;
		if (pd->port_rcvhdrtail_kvaddr) {
			dma_free_coherent(&dd->pcidev->dev, PAGE_SIZE,
					 (void *)pd->port_rcvhdrtail_kvaddr,
					 pd->port_rcvhdrqtailaddr_phys);
			pd->port_rcvhdrtail_kvaddr = NULL;
		}
	}
	if (pd->port_port && pd->port_rcvegrbuf) {
		unsigned e;

		for (e = 0; e < pd->port_rcvegrbuf_chunks; e++) {
			void *base = pd->port_rcvegrbuf[e];
			size_t size = pd->port_rcvegrbuf_size;

			ipath_cdbg(VERBOSE, "egrbuf free(%p, %lu), "
				   "chunk %u/%u\n", base,
				   (unsigned long) size,
				   e, pd->port_rcvegrbuf_chunks);
			dma_free_coherent(&dd->pcidev->dev, size,
				base, pd->port_rcvegrbuf_phys[e]);
		}
		vfree(pd->port_rcvegrbuf);
		pd->port_rcvegrbuf = NULL;
		vfree(pd->port_rcvegrbuf_phys);
		pd->port_rcvegrbuf_phys = NULL;
		pd->port_rcvegrbuf_chunks = 0;
	} else if (pd->port_port == 0 && dd->ipath_port0_skbs) {
		unsigned e;
		struct sk_buff **skbs = dd->ipath_port0_skbs;

		dd->ipath_port0_skbs = NULL;
		ipath_cdbg(VERBOSE, "free closed port %d ipath_port0_skbs "
			   "@ %p\n", pd->port_port, skbs);
		for (e = 0; e < dd->ipath_rcvegrcnt; e++)
			if (skbs[e])
				dev_kfree_skb(skbs[e]);
		vfree(skbs);
	}
	kfree(pd->port_tid_pg_list);
	kfree(pd);
}

static int __init infinipath_init(void)
{
	int ret;

	ipath_dbg(KERN_INFO DRIVER_LOAD_MSG "%s", ipath_core_version);

	/*
	 * These must be called before the driver is registered with
	 * the PCI subsystem.
	 */
	idr_init(&unit_table);
	if (!idr_pre_get(&unit_table, GFP_KERNEL)) {
		ret = -ENOMEM;
		goto bail;
	}

	ret = pci_register_driver(&ipath_driver);
	if (ret < 0) {
		printk(KERN_ERR IPATH_DRV_NAME
		       ": Unable to register driver: error %d\n", -ret);
		goto bail_unit;
	}

	ret = ipath_driver_create_group(&ipath_driver.driver);
	if (ret < 0) {
		printk(KERN_ERR IPATH_DRV_NAME ": Unable to create driver "
		       "sysfs entries: error %d\n", -ret);
		goto bail_pci;
	}

	ret = ipath_init_ipathfs();
	if (ret < 0) {
		printk(KERN_ERR IPATH_DRV_NAME ": Unable to create "
		       "ipathfs: error %d\n", -ret);
		goto bail_group;
	}

	goto bail;

bail_group:
	ipath_driver_remove_group(&ipath_driver.driver);

bail_pci:
	pci_unregister_driver(&ipath_driver);

bail_unit:
	idr_destroy(&unit_table);

bail:
	return ret;
}

static void cleanup_device(struct ipath_devdata *dd)
{
	int port;

	ipath_shutdown_device(dd);

	if (*dd->ipath_statusp & IPATH_STATUS_CHIP_PRESENT) {
		/* can't do anything more with chip; needs re-init */
		*dd->ipath_statusp &= ~IPATH_STATUS_CHIP_PRESENT;
		if (dd->ipath_kregbase) {
			/*
			 * if we haven't already cleaned up before these are
			 * to ensure any register reads/writes "fail" until
			 * re-init
			 */
			dd->ipath_kregbase = NULL;
			dd->ipath_uregbase = 0;
			dd->ipath_sregbase = 0;
			dd->ipath_cregbase = 0;
			dd->ipath_kregsize = 0;
		}
		ipath_disable_wc(dd);
	}

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

	if (dd->ipath_pageshadow) {
		struct page **tmpp = dd->ipath_pageshadow;
		int i, cnt = 0;

		ipath_cdbg(VERBOSE, "Unlocking any expTID pages still "
			   "locked\n");
		for (port = 0; port < dd->ipath_cfgports; port++) {
			int port_tidbase = port * dd->ipath_rcvtidcnt;
			int maxtid = port_tidbase + dd->ipath_rcvtidcnt;
			for (i = port_tidbase; i < maxtid; i++) {
				if (!tmpp[i])
					continue;
				ipath_release_user_pages(&tmpp[i], 1);
				tmpp[i] = NULL;
				cnt++;
			}
		}
		if (cnt) {
			ipath_stats.sps_pageunlocks += cnt;
			ipath_cdbg(VERBOSE, "There were still %u expTID "
				   "entries locked\n", cnt);
		}
		if (ipath_stats.sps_pagelocks ||
		    ipath_stats.sps_pageunlocks)
			ipath_cdbg(VERBOSE, "%llu pages locked, %llu "
				   "unlocked via ipath_m{un}lock\n",
				   (unsigned long long)
				   ipath_stats.sps_pagelocks,
				   (unsigned long long)
				   ipath_stats.sps_pageunlocks);

		ipath_cdbg(VERBOSE, "Free shadow page tid array at %p\n",
			   dd->ipath_pageshadow);
		vfree(dd->ipath_pageshadow);
		dd->ipath_pageshadow = NULL;
	}

	/*
	 * free any resources still in use (usually just kernel ports)
	 * at unload; we do for portcnt, not cfgports, because cfgports
	 * could have changed while we were loaded.
	 */
	for (port = 0; port < dd->ipath_portcnt; port++) {
		struct ipath_portdata *pd = dd->ipath_pd[port];
		dd->ipath_pd[port] = NULL;
		ipath_free_pddata(dd, pd);
	}
	kfree(dd->ipath_pd);
	/*
	 * debuggability, in case some cleanup path tries to use it
	 * after this
	 */
	dd->ipath_pd = NULL;
}

static void __exit infinipath_cleanup(void)
{
	struct ipath_devdata *dd, *tmp;
	unsigned long flags;

	ipath_exit_ipathfs();

	ipath_driver_remove_group(&ipath_driver.driver);

	spin_lock_irqsave(&ipath_devs_lock, flags);

	/*
	 * turn off rcv, send, and interrupts for all ports, all drivers
	 * should also hard reset the chip here?
	 * free up port 0 (kernel) rcvhdr, egr bufs, and eventually tid bufs
	 * for all versions of the driver, if they were allocated
	 */
	list_for_each_entry_safe(dd, tmp, &ipath_dev_list, ipath_list) {
		spin_unlock_irqrestore(&ipath_devs_lock, flags);

		if (dd->ipath_kregbase)
			cleanup_device(dd);

		if (dd->pcidev) {
			if (dd->pcidev->irq) {
				ipath_cdbg(VERBOSE,
					   "unit %u free_irq of irq %x\n",
					   dd->ipath_unit, dd->pcidev->irq);
				free_irq(dd->pcidev->irq, dd);
			} else
				ipath_dbg("irq is 0, not doing free_irq "
					  "for unit %u\n", dd->ipath_unit);

			/*
			 * 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->ipath_f_cleanup)
				/* clean up chip-specific stuff */
				dd->ipath_f_cleanup(dd);

			dd->pcidev = NULL;
		}
		spin_lock_irqsave(&ipath_devs_lock, flags);
	}

	spin_unlock_irqrestore(&ipath_devs_lock, flags);

	ipath_cdbg(VERBOSE, "Unregistering pci driver\n");
	pci_unregister_driver(&ipath_driver);

	idr_destroy(&unit_table);
}

/**
 * ipath_reset_device - reset the chip if possible
 * @unit: the device to reset
 *
 * Whether or not reset is successful, we attempt to re-initialize the chip
 * (that is, much like a driver unload/reload).  We clear the INITTED flag
 * so that the various entry points will fail until we reinitialize.  For
 * now, we only allow this if no user ports are open that use chip resources
 */
int ipath_reset_device(int unit)
{
	int ret, i;
	struct ipath_devdata *dd = ipath_lookup(unit);

	if (!dd) {
		ret = -ENODEV;
		goto bail;
	}

	dev_info(&dd->pcidev->dev, "Reset on unit %u requested\n", unit);

	if (!dd->ipath_kregbase || !(dd->ipath_flags & IPATH_PRESENT)) {
		dev_info(&dd->pcidev->dev, "Invalid unit number %u or "
			 "not initialized or not present\n", unit);
		ret = -ENXIO;
		goto bail;
	}

	if (dd->ipath_pd)
		for (i = 1; i < dd->ipath_cfgports; i++) {
			if (dd->ipath_pd[i] && dd->ipath_pd[i]->port_cnt) {
				ipath_dbg("unit %u port %d is in use "
					  "(PID %u cmd %s), can't reset\n",
					  unit, i,
					  dd->ipath_pd[i]->port_pid,
					  dd->ipath_pd[i]->port_comm);
				ret = -EBUSY;
				goto bail;
			}
		}

	dd->ipath_flags &= ~IPATH_INITTED;
	ret = dd->ipath_f_reset(dd);
	if (ret != 1)
		ipath_dbg("reset was not successful\n");
	ipath_dbg("Trying to reinitialize unit %u after reset attempt\n",
		  unit);
	ret = ipath_init_chip(dd, 1);
	if (ret)
		ipath_dev_err(dd, "Reinitialize unit %u after "
			      "reset failed with %d\n", unit, ret);
	else
		dev_info(&dd->pcidev->dev, "Reinitialized unit %u after "
			 "resetting\n", unit);

bail:
	return ret;
}

module_init(infinipath_init);
module_exit(infinipath_cleanup);