summaryrefslogtreecommitdiff
path: root/drivers/parisc/sba_iommu.c
blob: 48591badd4ceb243a02040cf9281e27cd2009cab (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
/*
**  System Bus Adapter (SBA) I/O MMU manager
**
**	(c) Copyright 2000-2004 Grant Grundler <grundler @ parisc-linux x org>
**	(c) Copyright 2004 Naresh Kumar Inna <knaresh at india x hp x com>
**	(c) Copyright 2000-2004 Hewlett-Packard Company
**
**	Portions (c) 1999 Dave S. Miller (from sparc64 I/O MMU code)
**
**	This program is free software; you can redistribute it and/or modify
**	it under the terms of the GNU General Public License as published by
**      the Free Software Foundation; either version 2 of the License, or
**      (at your option) any later version.
**
**
** This module initializes the IOC (I/O Controller) found on B1000/C3000/
** J5000/J7000/N-class/L-class machines and their successors.
**
** FIXME: add DMA hint support programming in both sba and lba modules.
*/

#include <linux/config.h>
#include <linux/types.h>
#include <linux/kernel.h>
#include <linux/spinlock.h>
#include <linux/slab.h>
#include <linux/init.h>

#include <linux/mm.h>
#include <linux/string.h>
#include <linux/pci.h>

#include <asm/byteorder.h>
#include <asm/io.h>
#include <asm/dma.h>		/* for DMA_CHUNK_SIZE */

#include <asm/hardware.h>	/* for register_parisc_driver() stuff */

#include <linux/proc_fs.h>
#include <asm/runway.h>		/* for proc_runway_root */
#include <asm/pdc.h>		/* for PDC_MODEL_* */
#include <asm/pdcpat.h>		/* for is_pdc_pat() */
#include <asm/parisc-device.h>


/* declared in arch/parisc/kernel/setup.c */
extern struct proc_dir_entry * proc_mckinley_root;

#define MODULE_NAME "SBA"

#ifdef CONFIG_PROC_FS
/* depends on proc fs support. But costs CPU performance */
#undef SBA_COLLECT_STATS
#endif

/*
** The number of debug flags is a clue - this code is fragile.
** Don't even think about messing with it unless you have
** plenty of 710's to sacrifice to the computer gods. :^)
*/
#undef DEBUG_SBA_INIT
#undef DEBUG_SBA_RUN
#undef DEBUG_SBA_RUN_SG
#undef DEBUG_SBA_RESOURCE
#undef ASSERT_PDIR_SANITY
#undef DEBUG_LARGE_SG_ENTRIES
#undef DEBUG_DMB_TRAP

#ifdef DEBUG_SBA_INIT
#define DBG_INIT(x...)	printk(x)
#else
#define DBG_INIT(x...)
#endif

#ifdef DEBUG_SBA_RUN
#define DBG_RUN(x...)	printk(x)
#else
#define DBG_RUN(x...)
#endif

#ifdef DEBUG_SBA_RUN_SG
#define DBG_RUN_SG(x...)	printk(x)
#else
#define DBG_RUN_SG(x...)
#endif


#ifdef DEBUG_SBA_RESOURCE
#define DBG_RES(x...)	printk(x)
#else
#define DBG_RES(x...)
#endif

#if defined(CONFIG_64BIT)
/* "low end" PA8800 machines use ZX1 chipset: PAT PDC and only run 64-bit */
#define ZX1_SUPPORT
#endif

#define SBA_INLINE	__inline__


/*
** The number of pdir entries to "free" before issueing
** a read to PCOM register to flush out PCOM writes.
** Interacts with allocation granularity (ie 4 or 8 entries
** allocated and free'd/purged at a time might make this
** less interesting).
*/
#define DELAYED_RESOURCE_CNT	16

#define DEFAULT_DMA_HINT_REG	0

#define ASTRO_RUNWAY_PORT	0x582
#define IKE_MERCED_PORT		0x803
#define REO_MERCED_PORT		0x804
#define REOG_MERCED_PORT	0x805
#define PLUTO_MCKINLEY_PORT	0x880

#define SBA_FUNC_ID	0x0000	/* function id */
#define SBA_FCLASS	0x0008	/* function class, bist, header, rev... */

#define IS_ASTRO(id)		((id)->hversion == ASTRO_RUNWAY_PORT)
#define IS_IKE(id)		((id)->hversion == IKE_MERCED_PORT)
#define IS_PLUTO(id)		((id)->hversion == PLUTO_MCKINLEY_PORT)

#define SBA_FUNC_SIZE 4096   /* SBA configuration function reg set */

#define ASTRO_IOC_OFFSET	(32 * SBA_FUNC_SIZE)
#define PLUTO_IOC_OFFSET	(1 * SBA_FUNC_SIZE)
/* Ike's IOC's occupy functions 2 and 3 */
#define IKE_IOC_OFFSET(p)	((p+2) * SBA_FUNC_SIZE)

#define IOC_CTRL          0x8	/* IOC_CTRL offset */
#define IOC_CTRL_TC       (1 << 0) /* TOC Enable */
#define IOC_CTRL_CE       (1 << 1) /* Coalesce Enable */
#define IOC_CTRL_DE       (1 << 2) /* Dillon Enable */
#define IOC_CTRL_RM       (1 << 8) /* Real Mode */
#define IOC_CTRL_NC       (1 << 9) /* Non Coherent Mode */
#define IOC_CTRL_D4       (1 << 11) /* Disable 4-byte coalescing */
#define IOC_CTRL_DD       (1 << 13) /* Disable distr. LMMIO range coalescing */

#define MAX_IOC		2	/* per Ike. Pluto/Astro only have 1. */

#define ROPES_PER_IOC	8	/* per Ike half or Pluto/Astro */


/*
** Offsets into MBIB (Function 0 on Ike and hopefully Astro)
** Firmware programs this stuff. Don't touch it.
*/
#define LMMIO_DIRECT0_BASE  0x300
#define LMMIO_DIRECT0_MASK  0x308
#define LMMIO_DIRECT0_ROUTE 0x310

#define LMMIO_DIST_BASE  0x360
#define LMMIO_DIST_MASK  0x368
#define LMMIO_DIST_ROUTE 0x370

#define IOS_DIST_BASE	0x390
#define IOS_DIST_MASK	0x398
#define IOS_DIST_ROUTE	0x3A0

#define IOS_DIRECT_BASE	0x3C0
#define IOS_DIRECT_MASK	0x3C8
#define IOS_DIRECT_ROUTE 0x3D0

/*
** Offsets into I/O TLB (Function 2 and 3 on Ike)
*/
#define ROPE0_CTL	0x200  /* "regbus pci0" */
#define ROPE1_CTL	0x208
#define ROPE2_CTL	0x210
#define ROPE3_CTL	0x218
#define ROPE4_CTL	0x220
#define ROPE5_CTL	0x228
#define ROPE6_CTL	0x230
#define ROPE7_CTL	0x238

#define HF_ENABLE	0x40


#define IOC_IBASE	0x300	/* IO TLB */
#define IOC_IMASK	0x308
#define IOC_PCOM	0x310
#define IOC_TCNFG	0x318
#define IOC_PDIR_BASE	0x320

/* AGP GART driver looks for this */
#define SBA_IOMMU_COOKIE    0x0000badbadc0ffeeUL


/*
** IOC supports 4/8/16/64KB page sizes (see TCNFG register)
** It's safer (avoid memory corruption) to keep DMA page mappings
** equivalently sized to VM PAGE_SIZE.
**
** We really can't avoid generating a new mapping for each
** page since the Virtual Coherence Index has to be generated
** and updated for each page.
**
** PAGE_SIZE could be greater than IOVP_SIZE. But not the inverse.
*/
#define IOVP_SIZE	PAGE_SIZE
#define IOVP_SHIFT	PAGE_SHIFT
#define IOVP_MASK	PAGE_MASK

#define SBA_PERF_CFG	0x708	/* Performance Counter stuff */
#define SBA_PERF_MASK1	0x718
#define SBA_PERF_MASK2	0x730


/*
** Offsets into PCI Performance Counters (functions 12 and 13)
** Controlled by PERF registers in function 2 & 3 respectively.
*/
#define SBA_PERF_CNT1	0x200
#define SBA_PERF_CNT2	0x208
#define SBA_PERF_CNT3	0x210


struct ioc {
	void __iomem	*ioc_hpa;	/* I/O MMU base address */
	char		*res_map;	/* resource map, bit == pdir entry */
	u64		*pdir_base;	/* physical base address */
	unsigned long	ibase;	/* pdir IOV Space base - shared w/lba_pci */
	unsigned long	imask;	/* pdir IOV Space mask - shared w/lba_pci */
#ifdef ZX1_SUPPORT
	unsigned long	iovp_mask;	/* help convert IOVA to IOVP */
#endif
	unsigned long	*res_hint;	/* next avail IOVP - circular search */
	spinlock_t	res_lock;
	unsigned int	res_bitshift;	/* from the LEFT! */
	unsigned int	res_size;	/* size of resource map in bytes */
#ifdef SBA_HINT_SUPPORT
/* FIXME : DMA HINTs not used */
	unsigned long	hint_mask_pdir;	/* bits used for DMA hints */
	unsigned int	hint_shift_pdir;
#endif
#if DELAYED_RESOURCE_CNT > 0
	int saved_cnt;
	struct sba_dma_pair {
		dma_addr_t	iova;
		size_t		size;
	} saved[DELAYED_RESOURCE_CNT];
#endif

#ifdef SBA_COLLECT_STATS
#define SBA_SEARCH_SAMPLE	0x100
	unsigned long avg_search[SBA_SEARCH_SAMPLE];
	unsigned long avg_idx;	/* current index into avg_search */
	unsigned long used_pages;
	unsigned long msingle_calls;
	unsigned long msingle_pages;
	unsigned long msg_calls;
	unsigned long msg_pages;
	unsigned long usingle_calls;
	unsigned long usingle_pages;
	unsigned long usg_calls;
	unsigned long usg_pages;
#endif

	/* STUFF We don't need in performance path */
	unsigned int	pdir_size;	/* in bytes, determined by IOV Space size */
};

struct sba_device {
	struct sba_device	*next;	/* list of SBA's in system */
	struct parisc_device	*dev;	/* dev found in bus walk */
	struct parisc_device_id	*iodc;	/* data about dev from firmware */
	const char 		*name;
	void __iomem		*sba_hpa; /* base address */
	spinlock_t		sba_lock;
	unsigned int		flags;  /* state/functionality enabled */
	unsigned int		hw_rev;  /* HW revision of chip */

	struct resource		chip_resv; /* MMIO reserved for chip */
	struct resource		iommu_resv; /* MMIO reserved for iommu */

	unsigned int		num_ioc;  /* number of on-board IOC's */
	struct ioc		ioc[MAX_IOC];
};


static struct sba_device *sba_list;

static unsigned long ioc_needs_fdc = 0;

/* global count of IOMMUs in the system */
static unsigned int global_ioc_cnt = 0;

/* PA8700 (Piranha 2.2) bug workaround */
static unsigned long piranha_bad_128k = 0;

/* Looks nice and keeps the compiler happy */
#define SBA_DEV(d) ((struct sba_device *) (d))

#ifdef SBA_AGP_SUPPORT
static int reserve_sba_gart = 1;
#endif

#define ROUNDUP(x,y) ((x + ((y)-1)) & ~((y)-1))


/************************************
** SBA register read and write support
**
** BE WARNED: register writes are posted.
**  (ie follow writes which must reach HW with a read)
**
** Superdome (in particular, REO) allows only 64-bit CSR accesses.
*/
#define READ_REG32(addr)	 le32_to_cpu(__raw_readl(addr))
#define READ_REG64(addr)	 le64_to_cpu(__raw_readq(addr))
#define WRITE_REG32(val, addr) __raw_writel(cpu_to_le32(val), addr)
#define WRITE_REG64(val, addr) __raw_writeq(cpu_to_le64(val), addr)

#ifdef CONFIG_64BIT
#define READ_REG(addr)		READ_REG64(addr)
#define WRITE_REG(value, addr)	WRITE_REG64(value, addr)
#else
#define READ_REG(addr)		READ_REG32(addr)
#define WRITE_REG(value, addr)	WRITE_REG32(value, addr)
#endif

#ifdef DEBUG_SBA_INIT

/* NOTE: When CONFIG_64BIT isn't defined, READ_REG64() is two 32-bit reads */

/**
 * sba_dump_ranges - debugging only - print ranges assigned to this IOA
 * @hpa: base address of the sba
 *
 * Print the MMIO and IO Port address ranges forwarded by an Astro/Ike/RIO
 * IO Adapter (aka Bus Converter).
 */
static void
sba_dump_ranges(void __iomem *hpa)
{
	DBG_INIT("SBA at 0x%p\n", hpa);
	DBG_INIT("IOS_DIST_BASE   : %Lx\n", READ_REG64(hpa+IOS_DIST_BASE));
	DBG_INIT("IOS_DIST_MASK   : %Lx\n", READ_REG64(hpa+IOS_DIST_MASK));
	DBG_INIT("IOS_DIST_ROUTE  : %Lx\n", READ_REG64(hpa+IOS_DIST_ROUTE));
	DBG_INIT("\n");
	DBG_INIT("IOS_DIRECT_BASE : %Lx\n", READ_REG64(hpa+IOS_DIRECT_BASE));
	DBG_INIT("IOS_DIRECT_MASK : %Lx\n", READ_REG64(hpa+IOS_DIRECT_MASK));
	DBG_INIT("IOS_DIRECT_ROUTE: %Lx\n", READ_REG64(hpa+IOS_DIRECT_ROUTE));
}

/**
 * sba_dump_tlb - debugging only - print IOMMU operating parameters
 * @hpa: base address of the IOMMU
 *
 * Print the size/location of the IO MMU PDIR.
 */
static void sba_dump_tlb(void __iomem *hpa)
{
	DBG_INIT("IO TLB at 0x%p\n", hpa);
	DBG_INIT("IOC_IBASE    : 0x%Lx\n", READ_REG64(hpa+IOC_IBASE));
	DBG_INIT("IOC_IMASK    : 0x%Lx\n", READ_REG64(hpa+IOC_IMASK));
	DBG_INIT("IOC_TCNFG    : 0x%Lx\n", READ_REG64(hpa+IOC_TCNFG));
	DBG_INIT("IOC_PDIR_BASE: 0x%Lx\n", READ_REG64(hpa+IOC_PDIR_BASE));
	DBG_INIT("\n");
}
#else
#define sba_dump_ranges(x)
#define sba_dump_tlb(x)
#endif	/* DEBUG_SBA_INIT */


#ifdef ASSERT_PDIR_SANITY

/**
 * sba_dump_pdir_entry - debugging only - print one IOMMU PDIR entry
 * @ioc: IO MMU structure which owns the pdir we are interested in.
 * @msg: text to print ont the output line.
 * @pide: pdir index.
 *
 * Print one entry of the IO MMU PDIR in human readable form.
 */
static void
sba_dump_pdir_entry(struct ioc *ioc, char *msg, uint pide)
{
	/* start printing from lowest pde in rval */
	u64 *ptr = &(ioc->pdir_base[pide & (~0U * BITS_PER_LONG)]);
	unsigned long *rptr = (unsigned long *) &(ioc->res_map[(pide >>3) & ~(sizeof(unsigned long) - 1)]);
	uint rcnt;

	printk(KERN_DEBUG "SBA: %s rp %p bit %d rval 0x%lx\n",
		 msg,
		 rptr, pide & (BITS_PER_LONG - 1), *rptr);

	rcnt = 0;
	while (rcnt < BITS_PER_LONG) {
		printk(KERN_DEBUG "%s %2d %p %016Lx\n",
			(rcnt == (pide & (BITS_PER_LONG - 1)))
				? "    -->" : "       ",
			rcnt, ptr, *ptr );
		rcnt++;
		ptr++;
	}
	printk(KERN_DEBUG "%s", msg);
}


/**
 * sba_check_pdir - debugging only - consistency checker
 * @ioc: IO MMU structure which owns the pdir we are interested in.
 * @msg: text to print ont the output line.
 *
 * Verify the resource map and pdir state is consistent
 */
static int
sba_check_pdir(struct ioc *ioc, char *msg)
{
	u32 *rptr_end = (u32 *) &(ioc->res_map[ioc->res_size]);
	u32 *rptr = (u32 *) ioc->res_map;	/* resource map ptr */
	u64 *pptr = ioc->pdir_base;	/* pdir ptr */
	uint pide = 0;

	while (rptr < rptr_end) {
		u32 rval = *rptr;
		int rcnt = 32;	/* number of bits we might check */

		while (rcnt) {
			/* Get last byte and highest bit from that */
			u32 pde = ((u32) (((char *)pptr)[7])) << 24;
			if ((rval ^ pde) & 0x80000000)
			{
				/*
				** BUMMER!  -- res_map != pdir --
				** Dump rval and matching pdir entries
				*/
				sba_dump_pdir_entry(ioc, msg, pide);
				return(1);
			}
			rcnt--;
			rval <<= 1;	/* try the next bit */
			pptr++;
			pide++;
		}
		rptr++;	/* look at next word of res_map */
	}
	/* It'd be nice if we always got here :^) */
	return 0;
}


/**
 * sba_dump_sg - debugging only - print Scatter-Gather list
 * @ioc: IO MMU structure which owns the pdir we are interested in.
 * @startsg: head of the SG list
 * @nents: number of entries in SG list
 *
 * print the SG list so we can verify it's correct by hand.
 */
static void
sba_dump_sg( struct ioc *ioc, struct scatterlist *startsg, int nents)
{
	while (nents-- > 0) {
		printk(KERN_DEBUG " %d : %08lx/%05x %p/%05x\n",
				nents,
				(unsigned long) sg_dma_address(startsg),
				sg_dma_len(startsg),
				sg_virt_addr(startsg), startsg->length);
		startsg++;
	}
}

#endif /* ASSERT_PDIR_SANITY */




/**************************************************************
*
*   I/O Pdir Resource Management
*
*   Bits set in the resource map are in use.
*   Each bit can represent a number of pages.
*   LSbs represent lower addresses (IOVA's).
*
***************************************************************/
#define PAGES_PER_RANGE 1	/* could increase this to 4 or 8 if needed */

/* Convert from IOVP to IOVA and vice versa. */

#ifdef ZX1_SUPPORT
/* Pluto (aka ZX1) boxes need to set or clear the ibase bits appropriately */
#define SBA_IOVA(ioc,iovp,offset,hint_reg) ((ioc->ibase) | (iovp) | (offset))
#define SBA_IOVP(ioc,iova) ((iova) & (ioc)->iovp_mask)
#else
/* only support Astro and ancestors. Saves a few cycles in key places */
#define SBA_IOVA(ioc,iovp,offset,hint_reg) ((iovp) | (offset))
#define SBA_IOVP(ioc,iova) (iova)
#endif

#define PDIR_INDEX(iovp)   ((iovp)>>IOVP_SHIFT)

#define RESMAP_MASK(n)    (~0UL << (BITS_PER_LONG - (n)))
#define RESMAP_IDX_MASK   (sizeof(unsigned long) - 1)


/**
 * sba_search_bitmap - find free space in IO PDIR resource bitmap
 * @ioc: IO MMU structure which owns the pdir we are interested in.
 * @bits_wanted: number of entries we need.
 *
 * Find consecutive free bits in resource bitmap.
 * Each bit represents one entry in the IO Pdir.
 * Cool perf optimization: search for log2(size) bits at a time.
 */
static SBA_INLINE unsigned long
sba_search_bitmap(struct ioc *ioc, unsigned long bits_wanted)
{
	unsigned long *res_ptr = ioc->res_hint;
	unsigned long *res_end = (unsigned long *) &(ioc->res_map[ioc->res_size]);
	unsigned long pide = ~0UL;

	if (bits_wanted > (BITS_PER_LONG/2)) {
		/* Search word at a time - no mask needed */
		for(; res_ptr < res_end; ++res_ptr) {
			if (*res_ptr == 0) {
				*res_ptr = RESMAP_MASK(bits_wanted);
				pide = ((unsigned long)res_ptr - (unsigned long)ioc->res_map);
				pide <<= 3;	/* convert to bit address */
				break;
			}
		}
		/* point to the next word on next pass */
		res_ptr++;
		ioc->res_bitshift = 0;
	} else {
		/*
		** Search the resource bit map on well-aligned values.
		** "o" is the alignment.
		** We need the alignment to invalidate I/O TLB using
		** SBA HW features in the unmap path.
		*/
		unsigned long o = 1 << get_order(bits_wanted << PAGE_SHIFT);
		uint bitshiftcnt = ROUNDUP(ioc->res_bitshift, o);
		unsigned long mask;

		if (bitshiftcnt >= BITS_PER_LONG) {
			bitshiftcnt = 0;
			res_ptr++;
		}
		mask = RESMAP_MASK(bits_wanted) >> bitshiftcnt;

		DBG_RES("%s() o %ld %p", __FUNCTION__, o, res_ptr);
		while(res_ptr < res_end)
		{ 
			DBG_RES("    %p %lx %lx\n", res_ptr, mask, *res_ptr);
			WARN_ON(mask == 0);
			if(((*res_ptr) & mask) == 0) {
				*res_ptr |= mask;     /* mark resources busy! */
				pide = ((unsigned long)res_ptr - (unsigned long)ioc->res_map);
				pide <<= 3;	/* convert to bit address */
				pide += bitshiftcnt;
				break;
			}
			mask >>= o;
			bitshiftcnt += o;
			if (mask == 0) {
				mask = RESMAP_MASK(bits_wanted);
				bitshiftcnt=0;
				res_ptr++;
			}
		}
		/* look in the same word on the next pass */
		ioc->res_bitshift = bitshiftcnt + bits_wanted;
	}

	/* wrapped ? */
	if (res_end <= res_ptr) {
		ioc->res_hint = (unsigned long *) ioc->res_map;
		ioc->res_bitshift = 0;
	} else {
		ioc->res_hint = res_ptr;
	}
	return (pide);
}


/**
 * sba_alloc_range - find free bits and mark them in IO PDIR resource bitmap
 * @ioc: IO MMU structure which owns the pdir we are interested in.
 * @size: number of bytes to create a mapping for
 *
 * Given a size, find consecutive unmarked and then mark those bits in the
 * resource bit map.
 */
static int
sba_alloc_range(struct ioc *ioc, size_t size)
{
	unsigned int pages_needed = size >> IOVP_SHIFT;
#ifdef SBA_COLLECT_STATS
	unsigned long cr_start = mfctl(16);
#endif
	unsigned long pide;

	pide = sba_search_bitmap(ioc, pages_needed);
	if (pide >= (ioc->res_size << 3)) {
		pide = sba_search_bitmap(ioc, pages_needed);
		if (pide >= (ioc->res_size << 3))
			panic("%s: I/O MMU @ %p is out of mapping resources\n",
			      __FILE__, ioc->ioc_hpa);
	}

#ifdef ASSERT_PDIR_SANITY
	/* verify the first enable bit is clear */
	if(0x00 != ((u8 *) ioc->pdir_base)[pide*sizeof(u64) + 7]) {
		sba_dump_pdir_entry(ioc, "sba_search_bitmap() botched it?", pide);
	}
#endif

	DBG_RES("%s(%x) %d -> %lx hint %x/%x\n",
		__FUNCTION__, size, pages_needed, pide,
		(uint) ((unsigned long) ioc->res_hint - (unsigned long) ioc->res_map),
		ioc->res_bitshift );

#ifdef SBA_COLLECT_STATS
	{
		unsigned long cr_end = mfctl(16);
		unsigned long tmp = cr_end - cr_start;
		/* check for roll over */
		cr_start = (cr_end < cr_start) ?  -(tmp) : (tmp);
	}
	ioc->avg_search[ioc->avg_idx++] = cr_start;
	ioc->avg_idx &= SBA_SEARCH_SAMPLE - 1;

	ioc->used_pages += pages_needed;
#endif

	return (pide);
}


/**
 * sba_free_range - unmark bits in IO PDIR resource bitmap
 * @ioc: IO MMU structure which owns the pdir we are interested in.
 * @iova: IO virtual address which was previously allocated.
 * @size: number of bytes to create a mapping for
 *
 * clear bits in the ioc's resource map
 */
static SBA_INLINE void
sba_free_range(struct ioc *ioc, dma_addr_t iova, size_t size)
{
	unsigned long iovp = SBA_IOVP(ioc, iova);
	unsigned int pide = PDIR_INDEX(iovp);
	unsigned int ridx = pide >> 3;	/* convert bit to byte address */
	unsigned long *res_ptr = (unsigned long *) &((ioc)->res_map[ridx & ~RESMAP_IDX_MASK]);

	int bits_not_wanted = size >> IOVP_SHIFT;

	/* 3-bits "bit" address plus 2 (or 3) bits for "byte" == bit in word */
	unsigned long m = RESMAP_MASK(bits_not_wanted) >> (pide & (BITS_PER_LONG - 1));

	DBG_RES("%s( ,%x,%x) %x/%lx %x %p %lx\n",
		__FUNCTION__, (uint) iova, size,
		bits_not_wanted, m, pide, res_ptr, *res_ptr);

#ifdef SBA_COLLECT_STATS
	ioc->used_pages -= bits_not_wanted;
#endif

	*res_ptr &= ~m;
}


/**************************************************************
*
*   "Dynamic DMA Mapping" support (aka "Coherent I/O")
*
***************************************************************/

#ifdef SBA_HINT_SUPPORT
#define SBA_DMA_HINT(ioc, val) ((val) << (ioc)->hint_shift_pdir)
#endif

typedef unsigned long space_t;
#define KERNEL_SPACE 0

/**
 * sba_io_pdir_entry - fill in one IO PDIR entry
 * @pdir_ptr:  pointer to IO PDIR entry
 * @sid: process Space ID - currently only support KERNEL_SPACE
 * @vba: Virtual CPU address of buffer to map
 * @hint: DMA hint set to use for this mapping
 *
 * SBA Mapping Routine
 *
 * Given a virtual address (vba, arg2) and space id, (sid, arg1)
 * sba_io_pdir_entry() loads the I/O PDIR entry pointed to by
 * pdir_ptr (arg0). 
 * Using the bass-ackwards HP bit numbering, Each IO Pdir entry
 * for Astro/Ike looks like:
 *
 *
 *  0                    19                                 51   55       63
 * +-+---------------------+----------------------------------+----+--------+
 * |V|        U            |            PPN[43:12]            | U  |   VI   |
 * +-+---------------------+----------------------------------+----+--------+
 *
 * Pluto is basically identical, supports fewer physical address bits:
 *
 *  0                       23                              51   55       63
 * +-+------------------------+-------------------------------+----+--------+
 * |V|        U               |         PPN[39:12]            | U  |   VI   |
 * +-+------------------------+-------------------------------+----+--------+
 *
 *  V  == Valid Bit  (Most Significant Bit is bit 0)
 *  U  == Unused
 * PPN == Physical Page Number
 * VI  == Virtual Index (aka Coherent Index)
 *
 * LPA instruction output is put into PPN field.
 * LCI (Load Coherence Index) instruction provides the "VI" bits.
 *
 * We pre-swap the bytes since PCX-W is Big Endian and the
 * IOMMU uses little endian for the pdir.
 */

void SBA_INLINE
sba_io_pdir_entry(u64 *pdir_ptr, space_t sid, unsigned long vba,
		  unsigned long hint)
{
	u64 pa; /* physical address */
	register unsigned ci; /* coherent index */

	pa = virt_to_phys(vba);
	pa &= IOVP_MASK;

	mtsp(sid,1);
	asm("lci 0(%%sr1, %1), %0" : "=r" (ci) : "r" (vba));
	pa |= (ci >> 12) & 0xff;  /* move CI (8 bits) into lowest byte */

	pa |= 0x8000000000000000ULL;	/* set "valid" bit */
	*pdir_ptr = cpu_to_le64(pa);	/* swap and store into I/O Pdir */

	/*
	 * If the PDC_MODEL capabilities has Non-coherent IO-PDIR bit set
	 * (bit #61, big endian), we have to flush and sync every time
	 * IO-PDIR is changed in Ike/Astro.
	 */
	if (ioc_needs_fdc)
		asm volatile("fdc %%r0(%0)" : : "r" (pdir_ptr));
}


/**
 * sba_mark_invalid - invalidate one or more IO PDIR entries
 * @ioc: IO MMU structure which owns the pdir we are interested in.
 * @iova:  IO Virtual Address mapped earlier
 * @byte_cnt:  number of bytes this mapping covers.
 *
 * Marking the IO PDIR entry(ies) as Invalid and invalidate
 * corresponding IO TLB entry. The Ike PCOM (Purge Command Register)
 * is to purge stale entries in the IO TLB when unmapping entries.
 *
 * The PCOM register supports purging of multiple pages, with a minium
 * of 1 page and a maximum of 2GB. Hardware requires the address be
 * aligned to the size of the range being purged. The size of the range
 * must be a power of 2. The "Cool perf optimization" in the
 * allocation routine helps keep that true.
 */
static SBA_INLINE void
sba_mark_invalid(struct ioc *ioc, dma_addr_t iova, size_t byte_cnt)
{
	u32 iovp = (u32) SBA_IOVP(ioc,iova);
	u64 *pdir_ptr = &ioc->pdir_base[PDIR_INDEX(iovp)];

#ifdef ASSERT_PDIR_SANITY
	/* Assert first pdir entry is set.
	**
	** Even though this is a big-endian machine, the entries
	** in the iopdir are little endian. That's why we look at
	** the byte at +7 instead of at +0.
	*/
	if (0x80 != (((u8 *) pdir_ptr)[7])) {
		sba_dump_pdir_entry(ioc,"sba_mark_invalid()", PDIR_INDEX(iovp));
	}
#endif

	if (byte_cnt > IOVP_SIZE)
	{
#if 0
		unsigned long entries_per_cacheline = ioc_needs_fdc ?
				L1_CACHE_ALIGN(((unsigned long) pdir_ptr))
					- (unsigned long) pdir_ptr;
				: 262144;
#endif

		/* set "size" field for PCOM */
		iovp |= get_order(byte_cnt) + PAGE_SHIFT;

		do {
			/* clear I/O Pdir entry "valid" bit first */
			((u8 *) pdir_ptr)[7] = 0;
			if (ioc_needs_fdc) {
				asm volatile("fdc %%r0(%0)" : : "r" (pdir_ptr));
#if 0
				entries_per_cacheline = L1_CACHE_SHIFT - 3;
#endif
			}
			pdir_ptr++;
			byte_cnt -= IOVP_SIZE;
		} while (byte_cnt > IOVP_SIZE);
	} else
		iovp |= IOVP_SHIFT;     /* set "size" field for PCOM */

	/*
	** clear I/O PDIR entry "valid" bit.
	** We have to R/M/W the cacheline regardless how much of the
	** pdir entry that we clobber.
	** The rest of the entry would be useful for debugging if we
	** could dump core on HPMC.
	*/
	((u8 *) pdir_ptr)[7] = 0;
	if (ioc_needs_fdc)
		asm volatile("fdc %%r0(%0)" : : "r" (pdir_ptr));

	WRITE_REG( SBA_IOVA(ioc, iovp, 0, 0), ioc->ioc_hpa+IOC_PCOM);
}

/**
 * sba_dma_supported - PCI driver can query DMA support
 * @dev: instance of PCI owned by the driver that's asking
 * @mask:  number of address bits this PCI device can handle
 *
 * See Documentation/DMA-mapping.txt
 */
static int sba_dma_supported( struct device *dev, u64 mask)
{
	struct ioc *ioc;

	if (dev == NULL) {
		printk(KERN_ERR MODULE_NAME ": EISA/ISA/et al not supported\n");
		BUG();
		return(0);
	}

	/* Documentation/DMA-mapping.txt tells drivers to try 64-bit first,
	 * then fall back to 32-bit if that fails.
	 * We are just "encouraging" 32-bit DMA masks here since we can
	 * never allow IOMMU bypass unless we add special support for ZX1.
	 */
	if (mask > ~0U)
		return 0;

	ioc = GET_IOC(dev);

	/*
	 * check if mask is >= than the current max IO Virt Address
	 * The max IO Virt address will *always* < 30 bits.
	 */
	return((int)(mask >= (ioc->ibase - 1 +
			(ioc->pdir_size / sizeof(u64) * IOVP_SIZE) )));
}


/**
 * sba_map_single - map one buffer and return IOVA for DMA
 * @dev: instance of PCI owned by the driver that's asking.
 * @addr:  driver buffer to map.
 * @size:  number of bytes to map in driver buffer.
 * @direction:  R/W or both.
 *
 * See Documentation/DMA-mapping.txt
 */
static dma_addr_t
sba_map_single(struct device *dev, void *addr, size_t size,
	       enum dma_data_direction direction)
{
	struct ioc *ioc;
	unsigned long flags; 
	dma_addr_t iovp;
	dma_addr_t offset;
	u64 *pdir_start;
	int pide;

	ioc = GET_IOC(dev);

	/* save offset bits */
	offset = ((dma_addr_t) (long) addr) & ~IOVP_MASK;

	/* round up to nearest IOVP_SIZE */
	size = (size + offset + ~IOVP_MASK) & IOVP_MASK;

	spin_lock_irqsave(&ioc->res_lock, flags);
#ifdef ASSERT_PDIR_SANITY
	sba_check_pdir(ioc,"Check before sba_map_single()");
#endif

#ifdef SBA_COLLECT_STATS
	ioc->msingle_calls++;
	ioc->msingle_pages += size >> IOVP_SHIFT;
#endif
	pide = sba_alloc_range(ioc, size);
	iovp = (dma_addr_t) pide << IOVP_SHIFT;

	DBG_RUN("%s() 0x%p -> 0x%lx\n",
		__FUNCTION__, addr, (long) iovp | offset);

	pdir_start = &(ioc->pdir_base[pide]);

	while (size > 0) {
		sba_io_pdir_entry(pdir_start, KERNEL_SPACE, (unsigned long) addr, 0);

		DBG_RUN("	pdir 0x%p %02x%02x%02x%02x%02x%02x%02x%02x\n",
			pdir_start,
			(u8) (((u8 *) pdir_start)[7]),
			(u8) (((u8 *) pdir_start)[6]),
			(u8) (((u8 *) pdir_start)[5]),
			(u8) (((u8 *) pdir_start)[4]),
			(u8) (((u8 *) pdir_start)[3]),
			(u8) (((u8 *) pdir_start)[2]),
			(u8) (((u8 *) pdir_start)[1]),
			(u8) (((u8 *) pdir_start)[0])
			);

		addr += IOVP_SIZE;
		size -= IOVP_SIZE;
		pdir_start++;
	}

	/* force FDC ops in io_pdir_entry() to be visible to IOMMU */
	if (ioc_needs_fdc)
		asm volatile("sync" : : );

#ifdef ASSERT_PDIR_SANITY
	sba_check_pdir(ioc,"Check after sba_map_single()");
#endif
	spin_unlock_irqrestore(&ioc->res_lock, flags);

	/* form complete address */
	return SBA_IOVA(ioc, iovp, offset, DEFAULT_DMA_HINT_REG);
}


/**
 * sba_unmap_single - unmap one IOVA and free resources
 * @dev: instance of PCI owned by the driver that's asking.
 * @iova:  IOVA of driver buffer previously mapped.
 * @size:  number of bytes mapped in driver buffer.
 * @direction:  R/W or both.
 *
 * See Documentation/DMA-mapping.txt
 */
static void
sba_unmap_single(struct device *dev, dma_addr_t iova, size_t size,
		 enum dma_data_direction direction)
{
	struct ioc *ioc;
#if DELAYED_RESOURCE_CNT > 0
	struct sba_dma_pair *d;
#endif
	unsigned long flags; 
	dma_addr_t offset;

	DBG_RUN("%s() iovp 0x%lx/%x\n", __FUNCTION__, (long) iova, size);

	ioc = GET_IOC(dev);
	offset = iova & ~IOVP_MASK;
	iova ^= offset;        /* clear offset bits */
	size += offset;
	size = ROUNDUP(size, IOVP_SIZE);

	spin_lock_irqsave(&ioc->res_lock, flags);

#ifdef SBA_COLLECT_STATS
	ioc->usingle_calls++;
	ioc->usingle_pages += size >> IOVP_SHIFT;
#endif

	sba_mark_invalid(ioc, iova, size);

#if DELAYED_RESOURCE_CNT > 0
	/* Delaying when we re-use a IO Pdir entry reduces the number
	 * of MMIO reads needed to flush writes to the PCOM register.
	 */
	d = &(ioc->saved[ioc->saved_cnt]);
	d->iova = iova;
	d->size = size;
	if (++(ioc->saved_cnt) >= DELAYED_RESOURCE_CNT) {
		int cnt = ioc->saved_cnt;
		while (cnt--) {
			sba_free_range(ioc, d->iova, d->size);
			d--;
		}
		ioc->saved_cnt = 0;

		READ_REG(ioc->ioc_hpa+IOC_PCOM);	/* flush purges */
	}
#else /* DELAYED_RESOURCE_CNT == 0 */
	sba_free_range(ioc, iova, size);

	/* If fdc's were issued, force fdc's to be visible now */
	if (ioc_needs_fdc)
		asm volatile("sync" : : );

	READ_REG(ioc->ioc_hpa+IOC_PCOM);	/* flush purges */
#endif /* DELAYED_RESOURCE_CNT == 0 */

	spin_unlock_irqrestore(&ioc->res_lock, flags);

	/* XXX REVISIT for 2.5 Linux - need syncdma for zero-copy support.
	** For Astro based systems this isn't a big deal WRT performance.
	** As long as 2.4 kernels copyin/copyout data from/to userspace,
	** we don't need the syncdma. The issue here is I/O MMU cachelines
	** are *not* coherent in all cases.  May be hwrev dependent.
	** Need to investigate more.
	asm volatile("syncdma");	
	*/
}


/**
 * sba_alloc_consistent - allocate/map shared mem for DMA
 * @hwdev: instance of PCI owned by the driver that's asking.
 * @size:  number of bytes mapped in driver buffer.
 * @dma_handle:  IOVA of new buffer.
 *
 * See Documentation/DMA-mapping.txt
 */
static void *sba_alloc_consistent(struct device *hwdev, size_t size,
					dma_addr_t *dma_handle, int gfp)
{
	void *ret;

	if (!hwdev) {
		/* only support PCI */
		*dma_handle = 0;
		return 0;
	}

        ret = (void *) __get_free_pages(gfp, get_order(size));

	if (ret) {
		memset(ret, 0, size);
		*dma_handle = sba_map_single(hwdev, ret, size, 0);
	}

	return ret;
}


/**
 * sba_free_consistent - free/unmap shared mem for DMA
 * @hwdev: instance of PCI owned by the driver that's asking.
 * @size:  number of bytes mapped in driver buffer.
 * @vaddr:  virtual address IOVA of "consistent" buffer.
 * @dma_handler:  IO virtual address of "consistent" buffer.
 *
 * See Documentation/DMA-mapping.txt
 */
static void
sba_free_consistent(struct device *hwdev, size_t size, void *vaddr,
		    dma_addr_t dma_handle)
{
	sba_unmap_single(hwdev, dma_handle, size, 0);
	free_pages((unsigned long) vaddr, get_order(size));
}


/*
** Since 0 is a valid pdir_base index value, can't use that
** to determine if a value is valid or not. Use a flag to indicate
** the SG list entry contains a valid pdir index.
*/
#define PIDE_FLAG 0x80000000UL

#ifdef SBA_COLLECT_STATS
#define IOMMU_MAP_STATS
#endif
#include "iommu-helpers.h"

#ifdef DEBUG_LARGE_SG_ENTRIES
int dump_run_sg = 0;
#endif


/**
 * sba_map_sg - map Scatter/Gather list
 * @dev: instance of PCI owned by the driver that's asking.
 * @sglist:  array of buffer/length pairs
 * @nents:  number of entries in list
 * @direction:  R/W or both.
 *
 * See Documentation/DMA-mapping.txt
 */
static int
sba_map_sg(struct device *dev, struct scatterlist *sglist, int nents,
	   enum dma_data_direction direction)
{
	struct ioc *ioc;
	int coalesced, filled = 0;
	unsigned long flags;

	DBG_RUN_SG("%s() START %d entries\n", __FUNCTION__, nents);

	ioc = GET_IOC(dev);

	/* Fast path single entry scatterlists. */
	if (nents == 1) {
		sg_dma_address(sglist) = sba_map_single(dev,
						(void *)sg_virt_addr(sglist),
						sglist->length, direction);
		sg_dma_len(sglist)     = sglist->length;
		return 1;
	}

	spin_lock_irqsave(&ioc->res_lock, flags);

#ifdef ASSERT_PDIR_SANITY
	if (sba_check_pdir(ioc,"Check before sba_map_sg()"))
	{
		sba_dump_sg(ioc, sglist, nents);
		panic("Check before sba_map_sg()");
	}
#endif

#ifdef SBA_COLLECT_STATS
	ioc->msg_calls++;
#endif

	/*
	** First coalesce the chunks and allocate I/O pdir space
	**
	** If this is one DMA stream, we can properly map using the
	** correct virtual address associated with each DMA page.
	** w/o this association, we wouldn't have coherent DMA!
	** Access to the virtual address is what forces a two pass algorithm.
	*/
	coalesced = iommu_coalesce_chunks(ioc, sglist, nents, sba_alloc_range);

	/*
	** Program the I/O Pdir
	**
	** map the virtual addresses to the I/O Pdir
	** o dma_address will contain the pdir index
	** o dma_len will contain the number of bytes to map 
	** o address contains the virtual address.
	*/
	filled = iommu_fill_pdir(ioc, sglist, nents, 0, sba_io_pdir_entry);

	/* force FDC ops in io_pdir_entry() to be visible to IOMMU */
	if (ioc_needs_fdc)
		asm volatile("sync" : : );

#ifdef ASSERT_PDIR_SANITY
	if (sba_check_pdir(ioc,"Check after sba_map_sg()"))
	{
		sba_dump_sg(ioc, sglist, nents);
		panic("Check after sba_map_sg()\n");
	}
#endif

	spin_unlock_irqrestore(&ioc->res_lock, flags);

	DBG_RUN_SG("%s() DONE %d mappings\n", __FUNCTION__, filled);

	return filled;
}


/**
 * sba_unmap_sg - unmap Scatter/Gather list
 * @dev: instance of PCI owned by the driver that's asking.
 * @sglist:  array of buffer/length pairs
 * @nents:  number of entries in list
 * @direction:  R/W or both.
 *
 * See Documentation/DMA-mapping.txt
 */
static void 
sba_unmap_sg(struct device *dev, struct scatterlist *sglist, int nents,
	     enum dma_data_direction direction)
{
	struct ioc *ioc;
#ifdef ASSERT_PDIR_SANITY
	unsigned long flags;
#endif

	DBG_RUN_SG("%s() START %d entries,  %p,%x\n",
		__FUNCTION__, nents, sg_virt_addr(sglist), sglist->length);

	ioc = GET_IOC(dev);

#ifdef SBA_COLLECT_STATS
	ioc->usg_calls++;
#endif

#ifdef ASSERT_PDIR_SANITY
	spin_lock_irqsave(&ioc->res_lock, flags);
	sba_check_pdir(ioc,"Check before sba_unmap_sg()");
	spin_unlock_irqrestore(&ioc->res_lock, flags);
#endif

	while (sg_dma_len(sglist) && nents--) {

		sba_unmap_single(dev, sg_dma_address(sglist), sg_dma_len(sglist), direction);
#ifdef SBA_COLLECT_STATS
		ioc->usg_pages += ((sg_dma_address(sglist) & ~IOVP_MASK) + sg_dma_len(sglist) + IOVP_SIZE - 1) >> PAGE_SHIFT;
		ioc->usingle_calls--;	/* kluge since call is unmap_sg() */
#endif
		++sglist;
	}

	DBG_RUN_SG("%s() DONE (nents %d)\n", __FUNCTION__,  nents);

#ifdef ASSERT_PDIR_SANITY
	spin_lock_irqsave(&ioc->res_lock, flags);
	sba_check_pdir(ioc,"Check after sba_unmap_sg()");
	spin_unlock_irqrestore(&ioc->res_lock, flags);
#endif

}

static struct hppa_dma_ops sba_ops = {
	.dma_supported =	sba_dma_supported,
	.alloc_consistent =	sba_alloc_consistent,
	.alloc_noncoherent =	sba_alloc_consistent,
	.free_consistent =	sba_free_consistent,
	.map_single =		sba_map_single,
	.unmap_single =		sba_unmap_single,
	.map_sg =		sba_map_sg,
	.unmap_sg =		sba_unmap_sg,
	.dma_sync_single_for_cpu =	NULL,
	.dma_sync_single_for_device =	NULL,
	.dma_sync_sg_for_cpu =		NULL,
	.dma_sync_sg_for_device =	NULL,
};


/**************************************************************************
**
**   SBA PAT PDC support
**
**   o call pdc_pat_cell_module()
**   o store ranges in PCI "resource" structures
**
**************************************************************************/

static void
sba_get_pat_resources(struct sba_device *sba_dev)
{
#if 0
/*
** TODO/REVISIT/FIXME: support for directed ranges requires calls to
**      PAT PDC to program the SBA/LBA directed range registers...this
**      burden may fall on the LBA code since it directly supports the
**      PCI subsystem. It's not clear yet. - ggg
*/
PAT_MOD(mod)->mod_info.mod_pages   = PAT_GET_MOD_PAGES(temp);
	FIXME : ???
PAT_MOD(mod)->mod_info.dvi         = PAT_GET_DVI(temp);
	Tells where the dvi bits are located in the address.
PAT_MOD(mod)->mod_info.ioc         = PAT_GET_IOC(temp);
	FIXME : ???
#endif
}


/**************************************************************
*
*   Initialization and claim
*
***************************************************************/
#define PIRANHA_ADDR_MASK	0x00160000UL /* bit 17,18,20 */
#define PIRANHA_ADDR_VAL	0x00060000UL /* bit 17,18 on */
static void *
sba_alloc_pdir(unsigned int pdir_size)
{
        unsigned long pdir_base;
	unsigned long pdir_order = get_order(pdir_size);

	pdir_base = __get_free_pages(GFP_KERNEL, pdir_order);
	if (NULL == (void *) pdir_base)	{
		panic("%s() could not allocate I/O Page Table\n",
			__FUNCTION__);
	}

	/* If this is not PA8700 (PCX-W2)
	**	OR newer than ver 2.2
	**	OR in a system that doesn't need VINDEX bits from SBA,
	**
	** then we aren't exposed to the HW bug.
	*/
	if ( ((boot_cpu_data.pdc.cpuid >> 5) & 0x7f) != 0x13
			|| (boot_cpu_data.pdc.versions > 0x202)
			|| (boot_cpu_data.pdc.capabilities & 0x08L) )
		return (void *) pdir_base;

	/*
	 * PA8700 (PCX-W2, aka piranha) silent data corruption fix
	 *
	 * An interaction between PA8700 CPU (Ver 2.2 or older) and
	 * Ike/Astro can cause silent data corruption. This is only
	 * a problem if the I/O PDIR is located in memory such that
	 * (little-endian)  bits 17 and 18 are on and bit 20 is off.
	 *
	 * Since the max IO Pdir size is 2MB, by cleverly allocating the
	 * right physical address, we can either avoid (IOPDIR <= 1MB)
	 * or minimize (2MB IO Pdir) the problem if we restrict the
	 * IO Pdir to a maximum size of 2MB-128K (1902K).
	 *
	 * Because we always allocate 2^N sized IO pdirs, either of the
	 * "bad" regions will be the last 128K if at all. That's easy
	 * to test for.
	 * 
	 */
	if (pdir_order <= (19-12)) {
		if (((virt_to_phys(pdir_base)+pdir_size-1) & PIRANHA_ADDR_MASK) == PIRANHA_ADDR_VAL) {
			/* allocate a new one on 512k alignment */
			unsigned long new_pdir = __get_free_pages(GFP_KERNEL, (19-12));
			/* release original */
			free_pages(pdir_base, pdir_order);

			pdir_base = new_pdir;

			/* release excess */
			while (pdir_order < (19-12)) {
				new_pdir += pdir_size;
				free_pages(new_pdir, pdir_order);
				pdir_order +=1;
				pdir_size <<=1;
			}
		}
	} else {
		/*
		** 1MB or 2MB Pdir
		** Needs to be aligned on an "odd" 1MB boundary.
		*/
		unsigned long new_pdir = __get_free_pages(GFP_KERNEL, pdir_order+1); /* 2 or 4MB */

		/* release original */
		free_pages( pdir_base, pdir_order);

		/* release first 1MB */
		free_pages(new_pdir, 20-12);

		pdir_base = new_pdir + 1024*1024;

		if (pdir_order > (20-12)) {
			/*
			** 2MB Pdir.
			**
			** Flag tells init_bitmap() to mark bad 128k as used
			** and to reduce the size by 128k.
			*/
			piranha_bad_128k = 1;

			new_pdir += 3*1024*1024;
			/* release last 1MB */
			free_pages(new_pdir, 20-12);

			/* release unusable 128KB */
			free_pages(new_pdir - 128*1024 , 17-12);

			pdir_size -= 128*1024;
		}
	}

	memset((void *) pdir_base, 0, pdir_size);
	return (void *) pdir_base;
}

static struct device *next_device(struct klist_iter *i)
{
        struct klist_node * n = klist_next(i);
        return n ? container_of(n, struct device, knode_parent) : NULL;
}

/* setup Mercury or Elroy IBASE/IMASK registers. */
static void 
setup_ibase_imask(struct parisc_device *sba, struct ioc *ioc, int ioc_num)
{
	/* lba_set_iregs() is in drivers/parisc/lba_pci.c */
        extern void lba_set_iregs(struct parisc_device *, u32, u32);
	struct device *dev;
	struct klist_iter i;

	klist_iter_init(&sba->dev.klist_children, &i);
	while ((dev = next_device(&i))) {
		struct parisc_device *lba = to_parisc_device(dev);
		int rope_num = (lba->hpa.start >> 13) & 0xf;
		if (rope_num >> 3 == ioc_num)
			lba_set_iregs(lba, ioc->ibase, ioc->imask);
	}
	klist_iter_exit(&i);
}

static void
sba_ioc_init_pluto(struct parisc_device *sba, struct ioc *ioc, int ioc_num)
{
	u32 iova_space_mask;
	u32 iova_space_size;
	int iov_order, tcnfg;
#ifdef SBA_AGP_SUPPORT
	int agp_found = 0;
#endif
	/*
	** Firmware programs the base and size of a "safe IOVA space"
	** (one that doesn't overlap memory or LMMIO space) in the
	** IBASE and IMASK registers.
	*/
	ioc->ibase = READ_REG(ioc->ioc_hpa + IOC_IBASE);
	iova_space_size = ~(READ_REG(ioc->ioc_hpa + IOC_IMASK) & 0xFFFFFFFFUL) + 1;

	if ((ioc->ibase < 0xfed00000UL) && ((ioc->ibase + iova_space_size) > 0xfee00000UL)) {
		printk("WARNING: IOV space overlaps local config and interrupt message, truncating\n");
		iova_space_size /= 2;
	}

	/*
	** iov_order is always based on a 1GB IOVA space since we want to
	** turn on the other half for AGP GART.
	*/
	iov_order = get_order(iova_space_size >> (IOVP_SHIFT - PAGE_SHIFT));
	ioc->pdir_size = (iova_space_size / IOVP_SIZE) * sizeof(u64);

	DBG_INIT("%s() hpa 0x%lx IOV %dMB (%d bits)\n",
		__FUNCTION__, ioc->ioc_hpa, iova_space_size >> 20,
		iov_order + PAGE_SHIFT);

	ioc->pdir_base = (void *) __get_free_pages(GFP_KERNEL,
						   get_order(ioc->pdir_size));
	if (!ioc->pdir_base)
		panic("Couldn't allocate I/O Page Table\n");

	memset(ioc->pdir_base, 0, ioc->pdir_size);

	DBG_INIT("%s() pdir %p size %x\n",
			__FUNCTION__, ioc->pdir_base, ioc->pdir_size);

#ifdef SBA_HINT_SUPPORT
	ioc->hint_shift_pdir = iov_order + PAGE_SHIFT;
	ioc->hint_mask_pdir = ~(0x3 << (iov_order + PAGE_SHIFT));

	DBG_INIT("	hint_shift_pdir %x hint_mask_pdir %lx\n",
		ioc->hint_shift_pdir, ioc->hint_mask_pdir);
#endif

	WARN_ON((((unsigned long) ioc->pdir_base) & PAGE_MASK) != (unsigned long) ioc->pdir_base);
	WRITE_REG(virt_to_phys(ioc->pdir_base), ioc->ioc_hpa + IOC_PDIR_BASE);

	/* build IMASK for IOC and Elroy */
	iova_space_mask =  0xffffffff;
	iova_space_mask <<= (iov_order + PAGE_SHIFT);
	ioc->imask = iova_space_mask;
#ifdef ZX1_SUPPORT
	ioc->iovp_mask = ~(iova_space_mask + PAGE_SIZE - 1);
#endif
	sba_dump_tlb(ioc->ioc_hpa);

	setup_ibase_imask(sba, ioc, ioc_num);

	WRITE_REG(ioc->imask, ioc->ioc_hpa + IOC_IMASK);

#ifdef CONFIG_64BIT
	/*
	** Setting the upper bits makes checking for bypass addresses
	** a little faster later on.
	*/
	ioc->imask |= 0xFFFFFFFF00000000UL;
#endif

	/* Set I/O PDIR Page size to system page size */
	switch (PAGE_SHIFT) {
		case 12: tcnfg = 0; break;	/*  4K */
		case 13: tcnfg = 1; break;	/*  8K */
		case 14: tcnfg = 2; break;	/* 16K */
		case 16: tcnfg = 3; break;	/* 64K */
		default:
			panic(__FILE__ "Unsupported system page size %d",
				1 << PAGE_SHIFT);
			break;
	}
	WRITE_REG(tcnfg, ioc->ioc_hpa + IOC_TCNFG);

	/*
	** Program the IOC's ibase and enable IOVA translation
	** Bit zero == enable bit.
	*/
	WRITE_REG(ioc->ibase | 1, ioc->ioc_hpa + IOC_IBASE);

	/*
	** Clear I/O TLB of any possible entries.
	** (Yes. This is a bit paranoid...but so what)
	*/
	WRITE_REG(ioc->ibase | 31, ioc->ioc_hpa + IOC_PCOM);

#ifdef SBA_AGP_SUPPORT
	/*
	** If an AGP device is present, only use half of the IOV space
	** for PCI DMA.  Unfortunately we can't know ahead of time
	** whether GART support will actually be used, for now we
	** can just key on any AGP device found in the system.
	** We program the next pdir index after we stop w/ a key for
	** the GART code to handshake on.
	*/
	device=NULL;
	for (lba = sba->child; lba; lba = lba->sibling) {
		if (IS_QUICKSILVER(lba))
			break;
	}

	if (lba) {
		DBG_INIT("%s: Reserving half of IOVA space for AGP GART support\n", __FUNCTION__);
		ioc->pdir_size /= 2;
		((u64 *)ioc->pdir_base)[PDIR_INDEX(iova_space_size/2)] = SBA_IOMMU_COOKIE;
	} else {
		DBG_INIT("%s: No GART needed - no AGP controller found\n", __FUNCTION__);
	}
#endif /* 0 */

}

static void
sba_ioc_init(struct parisc_device *sba, struct ioc *ioc, int ioc_num)
{
	u32 iova_space_size, iova_space_mask;
	unsigned int pdir_size, iov_order;

	/*
	** Determine IOVA Space size from memory size.
	**
	** Ideally, PCI drivers would register the maximum number
	** of DMA they can have outstanding for each device they
	** own.  Next best thing would be to guess how much DMA
	** can be outstanding based on PCI Class/sub-class. Both
	** methods still require some "extra" to support PCI
	** Hot-Plug/Removal of PCI cards. (aka PCI OLARD).
	**
	** While we have 32-bits "IOVA" space, top two 2 bits are used
	** for DMA hints - ergo only 30 bits max.
	*/

	iova_space_size = (u32) (num_physpages/global_ioc_cnt);

	/* limit IOVA space size to 1MB-1GB */
	if (iova_space_size < (1 << (20 - PAGE_SHIFT))) {
		iova_space_size = 1 << (20 - PAGE_SHIFT);
	}
	else if (iova_space_size > (1 << (30 - PAGE_SHIFT))) {
		iova_space_size = 1 << (30 - PAGE_SHIFT);
	}

	/*
	** iova space must be log2() in size.
	** thus, pdir/res_map will also be log2().
	** PIRANHA BUG: Exception is when IO Pdir is 2MB (gets reduced)
	*/
	iov_order = get_order(iova_space_size << PAGE_SHIFT);

	/* iova_space_size is now bytes, not pages */
	iova_space_size = 1 << (iov_order + PAGE_SHIFT);

	ioc->pdir_size = pdir_size = (iova_space_size/IOVP_SIZE) * sizeof(u64);

	DBG_INIT("%s() hpa 0x%lx mem %ldMB IOV %dMB (%d bits)\n",
			__FUNCTION__,
			ioc->ioc_hpa,
			(unsigned long) num_physpages >> (20 - PAGE_SHIFT),
			iova_space_size>>20,
			iov_order + PAGE_SHIFT);

	ioc->pdir_base = sba_alloc_pdir(pdir_size);

	DBG_INIT("%s() pdir %p size %x\n",
			__FUNCTION__, ioc->pdir_base, pdir_size);

#ifdef SBA_HINT_SUPPORT
	/* FIXME : DMA HINTs not used */
	ioc->hint_shift_pdir = iov_order + PAGE_SHIFT;
	ioc->hint_mask_pdir = ~(0x3 << (iov_order + PAGE_SHIFT));

	DBG_INIT("	hint_shift_pdir %x hint_mask_pdir %lx\n",
			ioc->hint_shift_pdir, ioc->hint_mask_pdir);
#endif

	WRITE_REG64(virt_to_phys(ioc->pdir_base), ioc->ioc_hpa + IOC_PDIR_BASE);

	/* build IMASK for IOC and Elroy */
	iova_space_mask =  0xffffffff;
	iova_space_mask <<= (iov_order + PAGE_SHIFT);

	/*
	** On C3000 w/512MB mem, HP-UX 10.20 reports:
	**     ibase=0, imask=0xFE000000, size=0x2000000.
	*/
	ioc->ibase = 0;
	ioc->imask = iova_space_mask;	/* save it */
#ifdef ZX1_SUPPORT
	ioc->iovp_mask = ~(iova_space_mask + PAGE_SIZE - 1);
#endif

	DBG_INIT("%s() IOV base 0x%lx mask 0x%0lx\n",
		__FUNCTION__, ioc->ibase, ioc->imask);

	/*
	** FIXME: Hint registers are programmed with default hint
	** values during boot, so hints should be sane even if we
	** can't reprogram them the way drivers want.
	*/

	setup_ibase_imask(sba, ioc, ioc_num);

	/*
	** Program the IOC's ibase and enable IOVA translation
	*/
	WRITE_REG(ioc->ibase | 1, ioc->ioc_hpa+IOC_IBASE);
	WRITE_REG(ioc->imask, ioc->ioc_hpa+IOC_IMASK);

	/* Set I/O PDIR Page size to 4K */
	WRITE_REG(0, ioc->ioc_hpa+IOC_TCNFG);

	/*
	** Clear I/O TLB of any possible entries.
	** (Yes. This is a bit paranoid...but so what)
	*/
	WRITE_REG(0 | 31, ioc->ioc_hpa+IOC_PCOM);

	ioc->ibase = 0; /* used by SBA_IOVA and related macros */	

	DBG_INIT("%s() DONE\n", __FUNCTION__);
}



/**************************************************************************
**
**   SBA initialization code (HW and SW)
**
**   o identify SBA chip itself
**   o initialize SBA chip modes (HardFail)
**   o initialize SBA chip modes (HardFail)
**   o FIXME: initialize DMA hints for reasonable defaults
**
**************************************************************************/

static void __iomem *ioc_remap(struct sba_device *sba_dev, int offset)
{
	return ioremap(sba_dev->dev->hpa.start + offset, SBA_FUNC_SIZE);
}

static void sba_hw_init(struct sba_device *sba_dev)
{ 
	int i;
	int num_ioc;
	u64 ioc_ctl;

	if (!is_pdc_pat()) {
		/* Shutdown the USB controller on Astro-based workstations.
		** Once we reprogram the IOMMU, the next DMA performed by
		** USB will HPMC the box. USB is only enabled if a
		** keyboard is present and found.
		**
		** With serial console, j6k v5.0 firmware says:
		**   mem_kbd hpa 0xfee003f8 sba 0x0 pad 0x0 cl_class 0x7
		**
		** FIXME: Using GFX+USB console at power up but direct
		**	linux to serial console is still broken.
		**	USB could generate DMA so we must reset USB.
		**	The proper sequence would be:
		**	o block console output
		**	o reset USB device
		**	o reprogram serial port
		**	o unblock console output
		*/
		if (PAGE0->mem_kbd.cl_class == CL_KEYBD) {
			pdc_io_reset_devices();
		}

	}


#if 0
printk("sba_hw_init(): mem_boot 0x%x 0x%x 0x%x 0x%x\n", PAGE0->mem_boot.hpa,
	PAGE0->mem_boot.spa, PAGE0->mem_boot.pad, PAGE0->mem_boot.cl_class);

	/*
	** Need to deal with DMA from LAN.
	**	Maybe use page zero boot device as a handle to talk
	**	to PDC about which device to shutdown.
	**
	** Netbooting, j6k v5.0 firmware says:
	** 	mem_boot hpa 0xf4008000 sba 0x0 pad 0x0 cl_class 0x1002
	** ARGH! invalid class.
	*/
	if ((PAGE0->mem_boot.cl_class != CL_RANDOM)
		&& (PAGE0->mem_boot.cl_class != CL_SEQU)) {
			pdc_io_reset();
	}
#endif

	if (!IS_PLUTO(sba_dev->iodc)) {
		ioc_ctl = READ_REG(sba_dev->sba_hpa+IOC_CTRL);
		DBG_INIT("%s() hpa 0x%lx ioc_ctl 0x%Lx ->",
			__FUNCTION__, sba_dev->sba_hpa, ioc_ctl);
		ioc_ctl &= ~(IOC_CTRL_RM | IOC_CTRL_NC | IOC_CTRL_CE);
		ioc_ctl |= IOC_CTRL_DD | IOC_CTRL_D4 | IOC_CTRL_TC;
			/* j6700 v1.6 firmware sets 0x294f */
			/* A500 firmware sets 0x4d */

		WRITE_REG(ioc_ctl, sba_dev->sba_hpa+IOC_CTRL);

#ifdef DEBUG_SBA_INIT
		ioc_ctl = READ_REG64(sba_dev->sba_hpa+IOC_CTRL);
		DBG_INIT(" 0x%Lx\n", ioc_ctl);
#endif
	} /* if !PLUTO */

	if (IS_ASTRO(sba_dev->iodc)) {
		int err;
		/* PAT_PDC (L-class) also reports the same goofy base */
		sba_dev->ioc[0].ioc_hpa = ioc_remap(sba_dev, ASTRO_IOC_OFFSET);
		num_ioc = 1;

		sba_dev->chip_resv.name = "Astro Intr Ack";
		sba_dev->chip_resv.start = PCI_F_EXTEND | 0xfef00000UL;
		sba_dev->chip_resv.end   = PCI_F_EXTEND | (0xff000000UL - 1) ;
		err = request_resource(&iomem_resource, &(sba_dev->chip_resv));
		if (err < 0) {
			BUG();
		}

	} else if (IS_PLUTO(sba_dev->iodc)) {
		int err;

		/* We use a negative value for IOC HPA so it gets 
                 * corrected when we add it with IKE's IOC offset.
		 * Doesnt look clean, but fewer code. 
                 */
		sba_dev->ioc[0].ioc_hpa = ioc_remap(sba_dev, PLUTO_IOC_OFFSET);
		num_ioc = 1;

		sba_dev->chip_resv.name = "Pluto Intr/PIOP/VGA";
		sba_dev->chip_resv.start = PCI_F_EXTEND | 0xfee00000UL;
		sba_dev->chip_resv.end   = PCI_F_EXTEND | (0xff200000UL - 1);
		err = request_resource(&iomem_resource, &(sba_dev->chip_resv));
		WARN_ON(err < 0);

		sba_dev->iommu_resv.name = "IOVA Space";
		sba_dev->iommu_resv.start = 0x40000000UL;
		sba_dev->iommu_resv.end   = 0x50000000UL - 1;
		err = request_resource(&iomem_resource, &(sba_dev->iommu_resv));
		WARN_ON(err < 0);
	} else {
		/* IS_IKE (ie N-class, L3000, L1500) */
		sba_dev->ioc[0].ioc_hpa = ioc_remap(sba_dev, IKE_IOC_OFFSET(0));
		sba_dev->ioc[1].ioc_hpa = ioc_remap(sba_dev, IKE_IOC_OFFSET(1));
		num_ioc = 2;

		/* TODO - LOOKUP Ike/Stretch chipset mem map */
	}
	/* XXX: What about Reo? */

	sba_dev->num_ioc = num_ioc;
	for (i = 0; i < num_ioc; i++) {
		/*
		** Make sure the box crashes if we get any errors on a rope.
		*/
		WRITE_REG(HF_ENABLE, sba_dev->ioc[i].ioc_hpa + ROPE0_CTL);
		WRITE_REG(HF_ENABLE, sba_dev->ioc[i].ioc_hpa + ROPE1_CTL);
		WRITE_REG(HF_ENABLE, sba_dev->ioc[i].ioc_hpa + ROPE2_CTL);
		WRITE_REG(HF_ENABLE, sba_dev->ioc[i].ioc_hpa + ROPE3_CTL);
		WRITE_REG(HF_ENABLE, sba_dev->ioc[i].ioc_hpa + ROPE4_CTL);
		WRITE_REG(HF_ENABLE, sba_dev->ioc[i].ioc_hpa + ROPE5_CTL);
		WRITE_REG(HF_ENABLE, sba_dev->ioc[i].ioc_hpa + ROPE6_CTL);
		WRITE_REG(HF_ENABLE, sba_dev->ioc[i].ioc_hpa + ROPE7_CTL);

		/* flush out the writes */
		READ_REG(sba_dev->ioc[i].ioc_hpa + ROPE7_CTL);

		DBG_INIT("	ioc[%d] ROPE_CFG 0x%Lx  ROPE_DBG 0x%Lx\n",
				i,
				READ_REG(sba_dev->ioc[i].ioc_hpa + 0x40),
				READ_REG(sba_dev->ioc[i].ioc_hpa + 0x50)
			);
		DBG_INIT("	STATUS_CONTROL 0x%Lx  FLUSH_CTRL 0x%Lx\n",
				READ_REG(sba_dev->ioc[i].ioc_hpa + 0x108),
				READ_REG(sba_dev->ioc[i].ioc_hpa + 0x400)
			);

		if (IS_PLUTO(sba_dev->iodc)) {
			sba_ioc_init_pluto(sba_dev->dev, &(sba_dev->ioc[i]), i);
		} else {
			sba_ioc_init(sba_dev->dev, &(sba_dev->ioc[i]), i);
		}
	}
}

static void
sba_common_init(struct sba_device *sba_dev)
{
	int i;

	/* add this one to the head of the list (order doesn't matter)
	** This will be useful for debugging - especially if we get coredumps
	*/
	sba_dev->next = sba_list;
	sba_list = sba_dev;

	for(i=0; i< sba_dev->num_ioc; i++) {
		int res_size;
#ifdef DEBUG_DMB_TRAP
		extern void iterate_pages(unsigned long , unsigned long ,
					  void (*)(pte_t * , unsigned long),
					  unsigned long );
		void set_data_memory_break(pte_t * , unsigned long);
#endif
		/* resource map size dictated by pdir_size */
		res_size = sba_dev->ioc[i].pdir_size/sizeof(u64); /* entries */

		/* Second part of PIRANHA BUG */
		if (piranha_bad_128k) {
			res_size -= (128*1024)/sizeof(u64);
		}

		res_size >>= 3;  /* convert bit count to byte count */
		DBG_INIT("%s() res_size 0x%x\n",
			__FUNCTION__, res_size);

		sba_dev->ioc[i].res_size = res_size;
		sba_dev->ioc[i].res_map = (char *) __get_free_pages(GFP_KERNEL, get_order(res_size));

#ifdef DEBUG_DMB_TRAP
		iterate_pages( sba_dev->ioc[i].res_map, res_size,
				set_data_memory_break, 0);
#endif

		if (NULL == sba_dev->ioc[i].res_map)
		{
			panic("%s:%s() could not allocate resource map\n",
			      __FILE__, __FUNCTION__ );
		}

		memset(sba_dev->ioc[i].res_map, 0, res_size);
		/* next available IOVP - circular search */
		sba_dev->ioc[i].res_hint = (unsigned long *)
				&(sba_dev->ioc[i].res_map[L1_CACHE_BYTES]);

#ifdef ASSERT_PDIR_SANITY
		/* Mark first bit busy - ie no IOVA 0 */
		sba_dev->ioc[i].res_map[0] = 0x80;
		sba_dev->ioc[i].pdir_base[0] = 0xeeffc0addbba0080ULL;
#endif

		/* Third (and last) part of PIRANHA BUG */
		if (piranha_bad_128k) {
			/* region from +1408K to +1536 is un-usable. */

			int idx_start = (1408*1024/sizeof(u64)) >> 3;
			int idx_end   = (1536*1024/sizeof(u64)) >> 3;
			long *p_start = (long *) &(sba_dev->ioc[i].res_map[idx_start]);
			long *p_end   = (long *) &(sba_dev->ioc[i].res_map[idx_end]);

			/* mark that part of the io pdir busy */
			while (p_start < p_end)
				*p_start++ = -1;
				
		}

#ifdef DEBUG_DMB_TRAP
		iterate_pages( sba_dev->ioc[i].res_map, res_size,
				set_data_memory_break, 0);
		iterate_pages( sba_dev->ioc[i].pdir_base, sba_dev->ioc[i].pdir_size,
				set_data_memory_break, 0);
#endif

		DBG_INIT("%s() %d res_map %x %p\n",
			__FUNCTION__, i, res_size, sba_dev->ioc[i].res_map);
	}

	spin_lock_init(&sba_dev->sba_lock);
	ioc_needs_fdc = boot_cpu_data.pdc.capabilities & PDC_MODEL_IOPDIR_FDC;

#ifdef DEBUG_SBA_INIT
	/*
	 * If the PDC_MODEL capabilities has Non-coherent IO-PDIR bit set
	 * (bit #61, big endian), we have to flush and sync every time
	 * IO-PDIR is changed in Ike/Astro.
	 */
	if (boot_cpu_data.pdc.capabilities & PDC_MODEL_IOPDIR_FDC) {
		printk(KERN_INFO MODULE_NAME " FDC/SYNC required.\n");
	} else {
		printk(KERN_INFO MODULE_NAME " IOC has cache coherent PDIR.\n");
	}
#endif
}

#ifdef CONFIG_PROC_FS
static int sba_proc_info(char *buf, char **start, off_t offset, int len)
{
	struct sba_device *sba_dev = sba_list;
	struct ioc *ioc = &sba_dev->ioc[0];	/* FIXME: Multi-IOC support! */
	int total_pages = (int) (ioc->res_size << 3); /* 8 bits per byte */
	unsigned long i;
#ifdef SBA_COLLECT_STATS
	unsigned long avg = 0, min, max;
#endif

	sprintf(buf, "%s rev %d.%d\n",
		sba_dev->name,
		(sba_dev->hw_rev & 0x7) + 1,
		(sba_dev->hw_rev & 0x18) >> 3
		);
	sprintf(buf, "%sIO PDIR size    : %d bytes (%d entries)\n",
		buf,
		(int) ((ioc->res_size << 3) * sizeof(u64)), /* 8 bits/byte */
		total_pages);

	sprintf(buf, "%sResource bitmap : %d bytes (%d pages)\n", 
		buf, ioc->res_size, ioc->res_size << 3);   /* 8 bits per byte */

	sprintf(buf, "%sLMMIO_BASE/MASK/ROUTE %08x %08x %08x\n",
		buf,
		READ_REG32(sba_dev->sba_hpa + LMMIO_DIST_BASE),
		READ_REG32(sba_dev->sba_hpa + LMMIO_DIST_MASK),
		READ_REG32(sba_dev->sba_hpa + LMMIO_DIST_ROUTE)
		);

	for (i=0; i<4; i++)
		sprintf(buf, "%sDIR%ld_BASE/MASK/ROUTE %08x %08x %08x\n",
			buf, i,
			READ_REG32(sba_dev->sba_hpa + LMMIO_DIRECT0_BASE  + i*0x18),
			READ_REG32(sba_dev->sba_hpa + LMMIO_DIRECT0_MASK  + i*0x18),
			READ_REG32(sba_dev->sba_hpa + LMMIO_DIRECT0_ROUTE + i*0x18)
		);

#ifdef SBA_COLLECT_STATS
	sprintf(buf, "%sIO PDIR entries : %ld free  %ld used (%d%%)\n", buf,
		total_pages - ioc->used_pages, ioc->used_pages,
		(int) (ioc->used_pages * 100 / total_pages));

	min = max = ioc->avg_search[0];
	for (i = 0; i < SBA_SEARCH_SAMPLE; i++) {
		avg += ioc->avg_search[i];
		if (ioc->avg_search[i] > max) max = ioc->avg_search[i];
		if (ioc->avg_search[i] < min) min = ioc->avg_search[i];
	}
	avg /= SBA_SEARCH_SAMPLE;
	sprintf(buf, "%s  Bitmap search : %ld/%ld/%ld (min/avg/max CPU Cycles)\n",
		buf, min, avg, max);

	sprintf(buf, "%spci_map_single(): %12ld calls  %12ld pages (avg %d/1000)\n",
		buf, ioc->msingle_calls, ioc->msingle_pages,
		(int) ((ioc->msingle_pages * 1000)/ioc->msingle_calls));

	/* KLUGE - unmap_sg calls unmap_single for each mapped page */
	min = ioc->usingle_calls;
	max = ioc->usingle_pages - ioc->usg_pages;
	sprintf(buf, "%spci_unmap_single: %12ld calls  %12ld pages (avg %d/1000)\n",
		buf, min, max,
		(int) ((max * 1000)/min));

	sprintf(buf, "%spci_map_sg()    : %12ld calls  %12ld pages (avg %d/1000)\n",
		buf, ioc->msg_calls, ioc->msg_pages,
		(int) ((ioc->msg_pages * 1000)/ioc->msg_calls));

	sprintf(buf, "%spci_unmap_sg()  : %12ld calls  %12ld pages (avg %d/1000)\n",
		buf, ioc->usg_calls, ioc->usg_pages,
		(int) ((ioc->usg_pages * 1000)/ioc->usg_calls));
#endif

	return strlen(buf);
}

#if 0
/* XXX too much output - exceeds 4k limit and needs to be re-written */
static int
sba_resource_map(char *buf, char **start, off_t offset, int len)
{
	struct sba_device *sba_dev = sba_list;
	struct ioc *ioc = &sba_dev->ioc[0];	/* FIXME: Mutli-IOC suppoer! */
	unsigned int *res_ptr = (unsigned int *)ioc->res_map;
	int i;

	buf[0] = '\0';
	for(i = 0; i < (ioc->res_size / sizeof(unsigned int)); ++i, ++res_ptr) {
		if ((i & 7) == 0)
		    strcat(buf,"\n   ");
		sprintf(buf, "%s %08x", buf, *res_ptr);
	}
	strcat(buf, "\n");

	return strlen(buf);
}
#endif /* 0 */
#endif /* CONFIG_PROC_FS */

static struct parisc_device_id sba_tbl[] = {
	{ HPHW_IOA, HVERSION_REV_ANY_ID, ASTRO_RUNWAY_PORT, 0xb },
	{ HPHW_BCPORT, HVERSION_REV_ANY_ID, IKE_MERCED_PORT, 0xc },
	{ HPHW_BCPORT, HVERSION_REV_ANY_ID, REO_MERCED_PORT, 0xc },
	{ HPHW_BCPORT, HVERSION_REV_ANY_ID, REOG_MERCED_PORT, 0xc },
	{ HPHW_IOA, HVERSION_REV_ANY_ID, PLUTO_MCKINLEY_PORT, 0xc },
	{ 0, }
};

int sba_driver_callback(struct parisc_device *);

static struct parisc_driver sba_driver = {
	.name =		MODULE_NAME,
	.id_table =	sba_tbl,
	.probe =	sba_driver_callback,
};

/*
** Determine if sba should claim this chip (return 0) or not (return 1).
** If so, initialize the chip and tell other partners in crime they
** have work to do.
*/
int
sba_driver_callback(struct parisc_device *dev)
{
	struct sba_device *sba_dev;
	u32 func_class;
	int i;
	char *version;
	void __iomem *sba_addr = ioremap(dev->hpa.start, SBA_FUNC_SIZE);

	sba_dump_ranges(sba_addr);

	/* Read HW Rev First */
	func_class = READ_REG(sba_addr + SBA_FCLASS);

	if (IS_ASTRO(&dev->id)) {
		unsigned long fclass;
		static char astro_rev[]="Astro ?.?";

		/* Astro is broken...Read HW Rev First */
		fclass = READ_REG(sba_addr);

		astro_rev[6] = '1' + (char) (fclass & 0x7);
		astro_rev[8] = '0' + (char) ((fclass & 0x18) >> 3);
		version = astro_rev;

	} else if (IS_IKE(&dev->id)) {
		static char ike_rev[] = "Ike rev ?";
		ike_rev[8] = '0' + (char) (func_class & 0xff);
		version = ike_rev;
	} else if (IS_PLUTO(&dev->id)) {
		static char pluto_rev[]="Pluto ?.?";
		pluto_rev[6] = '0' + (char) ((func_class & 0xf0) >> 4); 
		pluto_rev[8] = '0' + (char) (func_class & 0x0f); 
		version = pluto_rev;
	} else {
		static char reo_rev[] = "REO rev ?";
		reo_rev[8] = '0' + (char) (func_class & 0xff);
		version = reo_rev;
	}

	if (!global_ioc_cnt) {
		global_ioc_cnt = count_parisc_driver(&sba_driver);

		/* Astro and Pluto have one IOC per SBA */
		if ((!IS_ASTRO(&dev->id)) || (!IS_PLUTO(&dev->id)))
			global_ioc_cnt *= 2;
	}

	printk(KERN_INFO "%s found %s at 0x%lx\n",
		MODULE_NAME, version, dev->hpa.start);

	sba_dev = kmalloc(sizeof(struct sba_device), GFP_KERNEL);
	if (!sba_dev) {
		printk(KERN_ERR MODULE_NAME " - couldn't alloc sba_device\n");
		return -ENOMEM;
	}

	parisc_set_drvdata(dev, sba_dev);
	memset(sba_dev, 0, sizeof(struct sba_device));

	for(i=0; i<MAX_IOC; i++)
		spin_lock_init(&(sba_dev->ioc[i].res_lock));

	sba_dev->dev = dev;
	sba_dev->hw_rev = func_class;
	sba_dev->iodc = &dev->id;
	sba_dev->name = dev->name;
	sba_dev->sba_hpa = sba_addr;

	sba_get_pat_resources(sba_dev);
	sba_hw_init(sba_dev);
	sba_common_init(sba_dev);

	hppa_dma_ops = &sba_ops;

#ifdef CONFIG_PROC_FS
	if (IS_ASTRO(&dev->id)) {
		create_proc_info_entry("Astro", 0, proc_runway_root, sba_proc_info);
	} else if (IS_IKE(&dev->id)) {
		create_proc_info_entry("Ike", 0, proc_runway_root, sba_proc_info);
	} else if (IS_PLUTO(&dev->id)) {
		create_proc_info_entry("Pluto", 0, proc_mckinley_root, sba_proc_info);
	} else {
		create_proc_info_entry("Reo", 0, proc_runway_root, sba_proc_info);
	}
#if 0
	create_proc_info_entry("bitmap", 0, proc_runway_root, sba_resource_map);
#endif
#endif
	parisc_vmerge_boundary = IOVP_SIZE;
	parisc_vmerge_max_size = IOVP_SIZE * BITS_PER_LONG;
	parisc_has_iommu();
	return 0;
}

/*
** One time initialization to let the world know the SBA was found.
** This is the only routine which is NOT static.
** Must be called exactly once before pci_init().
*/
void __init sba_init(void)
{
	register_parisc_driver(&sba_driver);
}


/**
 * sba_get_iommu - Assign the iommu pointer for the pci bus controller.
 * @dev: The parisc device.
 *
 * Returns the appropriate IOMMU data for the given parisc PCI controller.
 * This is cached and used later for PCI DMA Mapping.
 */
void * sba_get_iommu(struct parisc_device *pci_hba)
{
	struct parisc_device *sba_dev = parisc_parent(pci_hba);
	struct sba_device *sba = sba_dev->dev.driver_data;
	char t = sba_dev->id.hw_type;
	int iocnum = (pci_hba->hw_path >> 3);	/* rope # */

	WARN_ON((t != HPHW_IOA) && (t != HPHW_BCPORT));

	return &(sba->ioc[iocnum]);
}


/**
 * sba_directed_lmmio - return first directed LMMIO range routed to rope
 * @pa_dev: The parisc device.
 * @r: resource PCI host controller wants start/end fields assigned.
 *
 * For the given parisc PCI controller, determine if any direct ranges
 * are routed down the corresponding rope.
 */
void sba_directed_lmmio(struct parisc_device *pci_hba, struct resource *r)
{
	struct parisc_device *sba_dev = parisc_parent(pci_hba);
	struct sba_device *sba = sba_dev->dev.driver_data;
	char t = sba_dev->id.hw_type;
	int i;
	int rope = (pci_hba->hw_path & (ROPES_PER_IOC-1));  /* rope # */

	if ((t!=HPHW_IOA) && (t!=HPHW_BCPORT))
		BUG();

	r->start = r->end = 0;

	/* Astro has 4 directed ranges. Not sure about Ike/Pluto/et al */
	for (i=0; i<4; i++) {
		int base, size;
		void __iomem *reg = sba->sba_hpa + i*0x18;

		base = READ_REG32(reg + LMMIO_DIRECT0_BASE);
		if ((base & 1) == 0)
			continue;	/* not enabled */

		size = READ_REG32(reg + LMMIO_DIRECT0_ROUTE);

		if ((size & (ROPES_PER_IOC-1)) != rope)
			continue;	/* directed down different rope */
		
		r->start = (base & ~1UL) | PCI_F_EXTEND;
		size = ~ READ_REG32(reg + LMMIO_DIRECT0_MASK);
		r->end = r->start + size;
	}
}


/**
 * sba_distributed_lmmio - return portion of distributed LMMIO range
 * @pa_dev: The parisc device.
 * @r: resource PCI host controller wants start/end fields assigned.
 *
 * For the given parisc PCI controller, return portion of distributed LMMIO
 * range. The distributed LMMIO is always present and it's just a question
 * of the base address and size of the range.
 */
void sba_distributed_lmmio(struct parisc_device *pci_hba, struct resource *r )
{
	struct parisc_device *sba_dev = parisc_parent(pci_hba);
	struct sba_device *sba = sba_dev->dev.driver_data;
	char t = sba_dev->id.hw_type;
	int base, size;
	int rope = (pci_hba->hw_path & (ROPES_PER_IOC-1));  /* rope # */

	if ((t!=HPHW_IOA) && (t!=HPHW_BCPORT))
		BUG();

	r->start = r->end = 0;

	base = READ_REG32(sba->sba_hpa + LMMIO_DIST_BASE);
	if ((base & 1) == 0) {
		BUG();	/* Gah! Distr Range wasn't enabled! */
		return;
	}

	r->start = (base & ~1UL) | PCI_F_EXTEND;

	size = (~READ_REG32(sba->sba_hpa + LMMIO_DIST_MASK)) / ROPES_PER_IOC;
	r->start += rope * (size + 1);	/* adjust base for this rope */
	r->end = r->start + size;
}