aboutsummaryrefslogtreecommitdiff
path: root/drivers/atm/ambassador.c
blob: bb3b016b6ce8efad07ea65d0df51cb0d6cf98d2d (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
/*
  Madge Ambassador ATM Adapter driver.
  Copyright (C) 1995-1999  Madge Networks Ltd.

  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 program is distributed in the hope that it will be useful,
  but WITHOUT ANY WARRANTY; without even the implied warranty of
  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  GNU General Public License for more details.

  You should have received a copy of the GNU General Public License
  along with this program; if not, write to the Free Software
  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA

  The GNU GPL is contained in /usr/doc/copyright/GPL on a Debian
  system and in the file COPYING in the Linux kernel source.
*/

/* * dedicated to the memory of Graham Gordon 1971-1998 * */

#include <linux/module.h>
#include <linux/types.h>
#include <linux/pci.h>
#include <linux/kernel.h>
#include <linux/init.h>
#include <linux/ioport.h>
#include <linux/atmdev.h>
#include <linux/delay.h>
#include <linux/interrupt.h>
#include <linux/poison.h>
#include <linux/bitrev.h>
#include <linux/mutex.h>
#include <linux/firmware.h>
#include <linux/ihex.h>
#include <linux/slab.h>

#include <asm/atomic.h>
#include <asm/io.h>
#include <asm/byteorder.h>

#include "ambassador.h"

#define maintainer_string "Giuliano Procida at Madge Networks <gprocida@madge.com>"
#define description_string "Madge ATM Ambassador driver"
#define version_string "1.2.4"

static inline void __init show_version (void) {
  printk ("%s version %s\n", description_string, version_string);
}

/*
  
  Theory of Operation
  
  I Hardware, detection, initialisation and shutdown.
  
  1. Supported Hardware
  
  This driver is for the PCI ATMizer-based Ambassador card (except
  very early versions). It is not suitable for the similar EISA "TR7"
  card. Commercially, both cards are known as Collage Server ATM
  adapters.
  
  The loader supports image transfer to the card, image start and few
  other miscellaneous commands.
  
  Only AAL5 is supported with vpi = 0 and vci in the range 0 to 1023.
  
  The cards are big-endian.
  
  2. Detection
  
  Standard PCI stuff, the early cards are detected and rejected.
  
  3. Initialisation
  
  The cards are reset and the self-test results are checked. The
  microcode image is then transferred and started. This waits for a
  pointer to a descriptor containing details of the host-based queues
  and buffers and various parameters etc. Once they are processed
  normal operations may begin. The BIA is read using a microcode
  command.
  
  4. Shutdown
  
  This may be accomplished either by a card reset or via the microcode
  shutdown command. Further investigation required.
  
  5. Persistent state
  
  The card reset does not affect PCI configuration (good) or the
  contents of several other "shared run-time registers" (bad) which
  include doorbell and interrupt control as well as EEPROM and PCI
  control. The driver must be careful when modifying these registers
  not to touch bits it does not use and to undo any changes at exit.
  
  II Driver software
  
  0. Generalities
  
  The adapter is quite intelligent (fast) and has a simple interface
  (few features). VPI is always zero, 1024 VCIs are supported. There
  is limited cell rate support. UBR channels can be capped and ABR
  (explicit rate, but not EFCI) is supported. There is no CBR or VBR
  support.
  
  1. Driver <-> Adapter Communication
  
  Apart from the basic loader commands, the driver communicates
  through three entities: the command queue (CQ), the transmit queue
  pair (TXQ) and the receive queue pairs (RXQ). These three entities
  are set up by the host and passed to the microcode just after it has
  been started.
  
  All queues are host-based circular queues. They are contiguous and
  (due to hardware limitations) have some restrictions as to their
  locations in (bus) memory. They are of the "full means the same as
  empty so don't do that" variety since the adapter uses pointers
  internally.
  
  The queue pairs work as follows: one queue is for supply to the
  adapter, items in it are pending and are owned by the adapter; the
  other is the queue for return from the adapter, items in it have
  been dealt with by the adapter. The host adds items to the supply
  (TX descriptors and free RX buffer descriptors) and removes items
  from the return (TX and RX completions). The adapter deals with out
  of order completions.
  
  Interrupts (card to host) and the doorbell (host to card) are used
  for signalling.
  
  1. CQ
  
  This is to communicate "open VC", "close VC", "get stats" etc. to
  the adapter. At most one command is retired every millisecond by the
  card. There is no out of order completion or notification. The
  driver needs to check the return code of the command, waiting as
  appropriate.
  
  2. TXQ
  
  TX supply items are of variable length (scatter gather support) and
  so the queue items are (more or less) pointers to the real thing.
  Each TX supply item contains a unique, host-supplied handle (the skb
  bus address seems most sensible as this works for Alphas as well,
  there is no need to do any endian conversions on the handles).
  
  TX return items consist of just the handles above.
  
  3. RXQ (up to 4 of these with different lengths and buffer sizes)
  
  RX supply items consist of a unique, host-supplied handle (the skb
  bus address again) and a pointer to the buffer data area.
  
  RX return items consist of the handle above, the VC, length and a
  status word. This just screams "oh so easy" doesn't it?

  Note on RX pool sizes:
   
  Each pool should have enough buffers to handle a back-to-back stream
  of minimum sized frames on a single VC. For example:
  
    frame spacing = 3us (about right)
    
    delay = IRQ lat + RX handling + RX buffer replenish = 20 (us)  (a guess)
    
    min number of buffers for one VC = 1 + delay/spacing (buffers)

    delay/spacing = latency = (20+2)/3 = 7 (buffers)  (rounding up)
    
  The 20us delay assumes that there is no need to sleep; if we need to
  sleep to get buffers we are going to drop frames anyway.
  
  In fact, each pool should have enough buffers to support the
  simultaneous reassembly of a separate frame on each VC and cope with
  the case in which frames complete in round robin cell fashion on
  each VC.
  
  Only one frame can complete at each cell arrival, so if "n" VCs are
  open, the worst case is to have them all complete frames together
  followed by all starting new frames together.
  
    desired number of buffers = n + delay/spacing
    
  These are the extreme requirements, however, they are "n+k" for some
  "k" so we have only the constant to choose. This is the argument
  rx_lats which current defaults to 7.
  
  Actually, "n ? n+k : 0" is better and this is what is implemented,
  subject to the limit given by the pool size.
  
  4. Driver locking
  
  Simple spinlocks are used around the TX and RX queue mechanisms.
  Anyone with a faster, working method is welcome to implement it.
  
  The adapter command queue is protected with a spinlock. We always
  wait for commands to complete.
  
  A more complex form of locking is used around parts of the VC open
  and close functions. There are three reasons for a lock: 1. we need
  to do atomic rate reservation and release (not used yet), 2. Opening
  sometimes involves two adapter commands which must not be separated
  by another command on the same VC, 3. the changes to RX pool size
  must be atomic. The lock needs to work over context switches, so we
  use a semaphore.
  
  III Hardware Features and Microcode Bugs
  
  1. Byte Ordering
  
  *%^"$&%^$*&^"$(%^$#&^%$(&#%$*(&^#%!"!"!*!
  
  2. Memory access
  
  All structures that are not accessed using DMA must be 4-byte
  aligned (not a problem) and must not cross 4MB boundaries.
  
  There is a DMA memory hole at E0000000-E00000FF (groan).
  
  TX fragments (DMA read) must not cross 4MB boundaries (would be 16MB
  but for a hardware bug).
  
  RX buffers (DMA write) must not cross 16MB boundaries and must
  include spare trailing bytes up to the next 4-byte boundary; they
  will be written with rubbish.
  
  The PLX likes to prefetch; if reading up to 4 u32 past the end of
  each TX fragment is not a problem, then TX can be made to go a
  little faster by passing a flag at init that disables a prefetch
  workaround. We do not pass this flag. (new microcode only)
  
  Now we:
  . Note that alloc_skb rounds up size to a 16byte boundary.  
  . Ensure all areas do not traverse 4MB boundaries.
  . Ensure all areas do not start at a E00000xx bus address.
  (I cannot be certain, but this may always hold with Linux)
  . Make all failures cause a loud message.
  . Discard non-conforming SKBs (causes TX failure or RX fill delay).
  . Discard non-conforming TX fragment descriptors (the TX fails).
  In the future we could:
  . Allow RX areas that traverse 4MB (but not 16MB) boundaries.
  . Segment TX areas into some/more fragments, when necessary.
  . Relax checks for non-DMA items (ignore hole).
  . Give scatter-gather (iovec) requirements using ???. (?)
  
  3. VC close is broken (only for new microcode)
  
  The VC close adapter microcode command fails to do anything if any
  frames have been received on the VC but none have been transmitted.
  Frames continue to be reassembled and passed (with IRQ) to the
  driver.
  
  IV To Do List
  
  . Fix bugs!
  
  . Timer code may be broken.
  
  . Deal with buggy VC close (somehow) in microcode 12.
  
  . Handle interrupted and/or non-blocking writes - is this a job for
    the protocol layer?
  
  . Add code to break up TX fragments when they span 4MB boundaries.
  
  . Add SUNI phy layer (need to know where SUNI lives on card).
  
  . Implement a tx_alloc fn to (a) satisfy TX alignment etc. and (b)
    leave extra headroom space for Ambassador TX descriptors.
  
  . Understand these elements of struct atm_vcc: recvq (proto?),
    sleep, callback, listenq, backlog_quota, reply and user_back.
  
  . Adjust TX/RX skb allocation to favour IP with LANE/CLIP (configurable).
  
  . Impose a TX-pending limit (2?) on each VC, help avoid TX q overflow.
  
  . Decide whether RX buffer recycling is or can be made completely safe;
    turn it back on. It looks like Werner is going to axe this.
  
  . Implement QoS changes on open VCs (involves extracting parts of VC open
    and close into separate functions and using them to make changes).
  
  . Hack on command queue so that someone can issue multiple commands and wait
    on the last one (OR only "no-op" or "wait" commands are waited for).
  
  . Eliminate need for while-schedule around do_command.
  
*/

static void do_housekeeping (unsigned long arg);
/********** globals **********/

static unsigned short debug = 0;
static unsigned int cmds = 8;
static unsigned int txs = 32;
static unsigned int rxs[NUM_RX_POOLS] = { 64, 64, 64, 64 };
static unsigned int rxs_bs[NUM_RX_POOLS] = { 4080, 12240, 36720, 65535 };
static unsigned int rx_lats = 7;
static unsigned char pci_lat = 0;

static const unsigned long onegigmask = -1 << 30;

/********** access to adapter **********/

static inline void wr_plain (const amb_dev * dev, size_t addr, u32 data) {
  PRINTD (DBG_FLOW|DBG_REGS, "wr: %08zx <- %08x", addr, data);
#ifdef AMB_MMIO
  dev->membase[addr / sizeof(u32)] = data;
#else
  outl (data, dev->iobase + addr);
#endif
}

static inline u32 rd_plain (const amb_dev * dev, size_t addr) {
#ifdef AMB_MMIO
  u32 data = dev->membase[addr / sizeof(u32)];
#else
  u32 data = inl (dev->iobase + addr);
#endif
  PRINTD (DBG_FLOW|DBG_REGS, "rd: %08zx -> %08x", addr, data);
  return data;
}

static inline void wr_mem (const amb_dev * dev, size_t addr, u32 data) {
  __be32 be = cpu_to_be32 (data);
  PRINTD (DBG_FLOW|DBG_REGS, "wr: %08zx <- %08x b[%08x]", addr, data, be);
#ifdef AMB_MMIO
  dev->membase[addr / sizeof(u32)] = be;
#else
  outl (be, dev->iobase + addr);
#endif
}

static inline u32 rd_mem (const amb_dev * dev, size_t addr) {
#ifdef AMB_MMIO
  __be32 be = dev->membase[addr / sizeof(u32)];
#else
  __be32 be = inl (dev->iobase + addr);
#endif
  u32 data = be32_to_cpu (be);
  PRINTD (DBG_FLOW|DBG_REGS, "rd: %08zx -> %08x b[%08x]", addr, data, be);
  return data;
}

/********** dump routines **********/

static inline void dump_registers (const amb_dev * dev) {
#ifdef DEBUG_AMBASSADOR
  if (debug & DBG_REGS) {
    size_t i;
    PRINTD (DBG_REGS, "reading PLX control: ");
    for (i = 0x00; i < 0x30; i += sizeof(u32))
      rd_mem (dev, i);
    PRINTD (DBG_REGS, "reading mailboxes: ");
    for (i = 0x40; i < 0x60; i += sizeof(u32))
      rd_mem (dev, i);
    PRINTD (DBG_REGS, "reading doorb irqev irqen reset:");
    for (i = 0x60; i < 0x70; i += sizeof(u32))
      rd_mem (dev, i);
  }
#else
  (void) dev;
#endif
  return;
}

static inline void dump_loader_block (volatile loader_block * lb) {
#ifdef DEBUG_AMBASSADOR
  unsigned int i;
  PRINTDB (DBG_LOAD, "lb @ %p; res: %d, cmd: %d, pay:",
	   lb, be32_to_cpu (lb->result), be32_to_cpu (lb->command));
  for (i = 0; i < MAX_COMMAND_DATA; ++i)
    PRINTDM (DBG_LOAD, " %08x", be32_to_cpu (lb->payload.data[i]));
  PRINTDE (DBG_LOAD, ", vld: %08x", be32_to_cpu (lb->valid));
#else
  (void) lb;
#endif
  return;
}

static inline void dump_command (command * cmd) {
#ifdef DEBUG_AMBASSADOR
  unsigned int i;
  PRINTDB (DBG_CMD, "cmd @ %p, req: %08x, pars:",
	   cmd, /*be32_to_cpu*/ (cmd->request));
  for (i = 0; i < 3; ++i)
    PRINTDM (DBG_CMD, " %08x", /*be32_to_cpu*/ (cmd->args.par[i]));
  PRINTDE (DBG_CMD, "");
#else
  (void) cmd;
#endif
  return;
}

static inline void dump_skb (char * prefix, unsigned int vc, struct sk_buff * skb) {
#ifdef DEBUG_AMBASSADOR
  unsigned int i;
  unsigned char * data = skb->data;
  PRINTDB (DBG_DATA, "%s(%u) ", prefix, vc);
  for (i=0; i<skb->len && i < 256;i++)
    PRINTDM (DBG_DATA, "%02x ", data[i]);
  PRINTDE (DBG_DATA,"");
#else
  (void) prefix;
  (void) vc;
  (void) skb;
#endif
  return;
}

/********** check memory areas for use by Ambassador **********/

/* see limitations under Hardware Features */

static int check_area (void * start, size_t length) {
  // assumes length > 0
  const u32 fourmegmask = -1 << 22;
  const u32 twofivesixmask = -1 << 8;
  const u32 starthole = 0xE0000000;
  u32 startaddress = virt_to_bus (start);
  u32 lastaddress = startaddress+length-1;
  if ((startaddress ^ lastaddress) & fourmegmask ||
      (startaddress & twofivesixmask) == starthole) {
    PRINTK (KERN_ERR, "check_area failure: [%x,%x] - mail maintainer!",
	    startaddress, lastaddress);
    return -1;
  } else {
    return 0;
  }
}

/********** free an skb (as per ATM device driver documentation) **********/

static void amb_kfree_skb (struct sk_buff * skb) {
  if (ATM_SKB(skb)->vcc->pop) {
    ATM_SKB(skb)->vcc->pop (ATM_SKB(skb)->vcc, skb);
  } else {
    dev_kfree_skb_any (skb);
  }
}

/********** TX completion **********/

static void tx_complete (amb_dev * dev, tx_out * tx) {
  tx_simple * tx_descr = bus_to_virt (tx->handle);
  struct sk_buff * skb = tx_descr->skb;
  
  PRINTD (DBG_FLOW|DBG_TX, "tx_complete %p %p", dev, tx);
  
  // VC layer stats
  atomic_inc(&ATM_SKB(skb)->vcc->stats->tx);
  
  // free the descriptor
  kfree (tx_descr);
  
  // free the skb
  amb_kfree_skb (skb);
  
  dev->stats.tx_ok++;
  return;
}

/********** RX completion **********/

static void rx_complete (amb_dev * dev, rx_out * rx) {
  struct sk_buff * skb = bus_to_virt (rx->handle);
  u16 vc = be16_to_cpu (rx->vc);
  // unused: u16 lec_id = be16_to_cpu (rx->lec_id);
  u16 status = be16_to_cpu (rx->status);
  u16 rx_len = be16_to_cpu (rx->length);
  
  PRINTD (DBG_FLOW|DBG_RX, "rx_complete %p %p (len=%hu)", dev, rx, rx_len);
  
  // XXX move this in and add to VC stats ???
  if (!status) {
    struct atm_vcc * atm_vcc = dev->rxer[vc];
    dev->stats.rx.ok++;
    
    if (atm_vcc) {
      
      if (rx_len <= atm_vcc->qos.rxtp.max_sdu) {
	
	if (atm_charge (atm_vcc, skb->truesize)) {
	  
	  // prepare socket buffer
	  ATM_SKB(skb)->vcc = atm_vcc;
	  skb_put (skb, rx_len);
	  
	  dump_skb ("<<<", vc, skb);
	  
	  // VC layer stats
	  atomic_inc(&atm_vcc->stats->rx);
	  __net_timestamp(skb);
	  // end of our responsibility
	  atm_vcc->push (atm_vcc, skb);
	  return;
	  
	} else {
	  // someone fix this (message), please!
	  PRINTD (DBG_INFO|DBG_RX, "dropped thanks to atm_charge (vc %hu, truesize %u)", vc, skb->truesize);
	  // drop stats incremented in atm_charge
	}
	
      } else {
      	PRINTK (KERN_INFO, "dropped over-size frame");
	// should we count this?
	atomic_inc(&atm_vcc->stats->rx_drop);
      }
      
    } else {
      PRINTD (DBG_WARN|DBG_RX, "got frame but RX closed for channel %hu", vc);
      // this is an adapter bug, only in new version of microcode
    }
    
  } else {
    dev->stats.rx.error++;
    if (status & CRC_ERR)
      dev->stats.rx.badcrc++;
    if (status & LEN_ERR)
      dev->stats.rx.toolong++;
    if (status & ABORT_ERR)
      dev->stats.rx.aborted++;
    if (status & UNUSED_ERR)
      dev->stats.rx.unused++;
  }
  
  dev_kfree_skb_any (skb);
  return;
}

/*
  
  Note on queue handling.
  
  Here "give" and "take" refer to queue entries and a queue (pair)
  rather than frames to or from the host or adapter. Empty frame
  buffers are given to the RX queue pair and returned unused or
  containing RX frames. TX frames (well, pointers to TX fragment
  lists) are given to the TX queue pair, completions are returned.
  
*/

/********** command queue **********/

// I really don't like this, but it's the best I can do at the moment

// also, the callers are responsible for byte order as the microcode
// sometimes does 16-bit accesses (yuk yuk yuk)

static int command_do (amb_dev * dev, command * cmd) {
  amb_cq * cq = &dev->cq;
  volatile amb_cq_ptrs * ptrs = &cq->ptrs;
  command * my_slot;
  
  PRINTD (DBG_FLOW|DBG_CMD, "command_do %p", dev);
  
  if (test_bit (dead, &dev->flags))
    return 0;
  
  spin_lock (&cq->lock);
  
  // if not full...
  if (cq->pending < cq->maximum) {
    // remember my slot for later
    my_slot = ptrs->in;
    PRINTD (DBG_CMD, "command in slot %p", my_slot);
    
    dump_command (cmd);
    
    // copy command in
    *ptrs->in = *cmd;
    cq->pending++;
    ptrs->in = NEXTQ (ptrs->in, ptrs->start, ptrs->limit);
    
    // mail the command
    wr_mem (dev, offsetof(amb_mem, mb.adapter.cmd_address), virt_to_bus (ptrs->in));
    
    if (cq->pending > cq->high)
      cq->high = cq->pending;
    spin_unlock (&cq->lock);
    
    // these comments were in a while-loop before, msleep removes the loop
    // go to sleep
    // PRINTD (DBG_CMD, "wait: sleeping %lu for command", timeout);
    msleep(cq->pending);
    
    // wait for my slot to be reached (all waiters are here or above, until...)
    while (ptrs->out != my_slot) {
      PRINTD (DBG_CMD, "wait: command slot (now at %p)", ptrs->out);
      set_current_state(TASK_UNINTERRUPTIBLE);
      schedule();
    }
    
    // wait on my slot (... one gets to its slot, and... )
    while (ptrs->out->request != cpu_to_be32 (SRB_COMPLETE)) {
      PRINTD (DBG_CMD, "wait: command slot completion");
      set_current_state(TASK_UNINTERRUPTIBLE);
      schedule();
    }
    
    PRINTD (DBG_CMD, "command complete");
    // update queue (... moves the queue along to the next slot)
    spin_lock (&cq->lock);
    cq->pending--;
    // copy command out
    *cmd = *ptrs->out;
    ptrs->out = NEXTQ (ptrs->out, ptrs->start, ptrs->limit);
    spin_unlock (&cq->lock);
    
    return 0;
  } else {
    cq->filled++;
    spin_unlock (&cq->lock);
    return -EAGAIN;
  }
  
}

/********** TX queue pair **********/

static int tx_give (amb_dev * dev, tx_in * tx) {
  amb_txq * txq = &dev->txq;
  unsigned long flags;
  
  PRINTD (DBG_FLOW|DBG_TX, "tx_give %p", dev);

  if (test_bit (dead, &dev->flags))
    return 0;
  
  spin_lock_irqsave (&txq->lock, flags);
  
  if (txq->pending < txq->maximum) {
    PRINTD (DBG_TX, "TX in slot %p", txq->in.ptr);

    *txq->in.ptr = *tx;
    txq->pending++;
    txq->in.ptr = NEXTQ (txq->in.ptr, txq->in.start, txq->in.limit);
    // hand over the TX and ring the bell
    wr_mem (dev, offsetof(amb_mem, mb.adapter.tx_address), virt_to_bus (txq->in.ptr));
    wr_mem (dev, offsetof(amb_mem, doorbell), TX_FRAME);
    
    if (txq->pending > txq->high)
      txq->high = txq->pending;
    spin_unlock_irqrestore (&txq->lock, flags);
    return 0;
  } else {
    txq->filled++;
    spin_unlock_irqrestore (&txq->lock, flags);
    return -EAGAIN;
  }
}

static int tx_take (amb_dev * dev) {
  amb_txq * txq = &dev->txq;
  unsigned long flags;
  
  PRINTD (DBG_FLOW|DBG_TX, "tx_take %p", dev);
  
  spin_lock_irqsave (&txq->lock, flags);
  
  if (txq->pending && txq->out.ptr->handle) {
    // deal with TX completion
    tx_complete (dev, txq->out.ptr);
    // mark unused again
    txq->out.ptr->handle = 0;
    // remove item
    txq->pending--;
    txq->out.ptr = NEXTQ (txq->out.ptr, txq->out.start, txq->out.limit);
    
    spin_unlock_irqrestore (&txq->lock, flags);
    return 0;
  } else {
    
    spin_unlock_irqrestore (&txq->lock, flags);
    return -1;
  }
}

/********** RX queue pairs **********/

static int rx_give (amb_dev * dev, rx_in * rx, unsigned char pool) {
  amb_rxq * rxq = &dev->rxq[pool];
  unsigned long flags;
  
  PRINTD (DBG_FLOW|DBG_RX, "rx_give %p[%hu]", dev, pool);
  
  spin_lock_irqsave (&rxq->lock, flags);
  
  if (rxq->pending < rxq->maximum) {
    PRINTD (DBG_RX, "RX in slot %p", rxq->in.ptr);

    *rxq->in.ptr = *rx;
    rxq->pending++;
    rxq->in.ptr = NEXTQ (rxq->in.ptr, rxq->in.start, rxq->in.limit);
    // hand over the RX buffer
    wr_mem (dev, offsetof(amb_mem, mb.adapter.rx_address[pool]), virt_to_bus (rxq->in.ptr));
    
    spin_unlock_irqrestore (&rxq->lock, flags);
    return 0;
  } else {
    spin_unlock_irqrestore (&rxq->lock, flags);
    return -1;
  }
}

static int rx_take (amb_dev * dev, unsigned char pool) {
  amb_rxq * rxq = &dev->rxq[pool];
  unsigned long flags;
  
  PRINTD (DBG_FLOW|DBG_RX, "rx_take %p[%hu]", dev, pool);
  
  spin_lock_irqsave (&rxq->lock, flags);
  
  if (rxq->pending && (rxq->out.ptr->status || rxq->out.ptr->length)) {
    // deal with RX completion
    rx_complete (dev, rxq->out.ptr);
    // mark unused again
    rxq->out.ptr->status = 0;
    rxq->out.ptr->length = 0;
    // remove item
    rxq->pending--;
    rxq->out.ptr = NEXTQ (rxq->out.ptr, rxq->out.start, rxq->out.limit);
    
    if (rxq->pending < rxq->low)
      rxq->low = rxq->pending;
    spin_unlock_irqrestore (&rxq->lock, flags);
    return 0;
  } else {
    if (!rxq->pending && rxq->buffers_wanted)
      rxq->emptied++;
    spin_unlock_irqrestore (&rxq->lock, flags);
    return -1;
  }
}

/********** RX Pool handling **********/

/* pre: buffers_wanted = 0, post: pending = 0 */
static void drain_rx_pool (amb_dev * dev, unsigned char pool) {
  amb_rxq * rxq = &dev->rxq[pool];
  
  PRINTD (DBG_FLOW|DBG_POOL, "drain_rx_pool %p %hu", dev, pool);
  
  if (test_bit (dead, &dev->flags))
    return;
  
  /* we are not quite like the fill pool routines as we cannot just
     remove one buffer, we have to remove all of them, but we might as
     well pretend... */
  if (rxq->pending > rxq->buffers_wanted) {
    command cmd;
    cmd.request = cpu_to_be32 (SRB_FLUSH_BUFFER_Q);
    cmd.args.flush.flags = cpu_to_be32 (pool << SRB_POOL_SHIFT);
    while (command_do (dev, &cmd))
      schedule();
    /* the pool may also be emptied via the interrupt handler */
    while (rxq->pending > rxq->buffers_wanted)
      if (rx_take (dev, pool))
	schedule();
  }
  
  return;
}

static void drain_rx_pools (amb_dev * dev) {
  unsigned char pool;
  
  PRINTD (DBG_FLOW|DBG_POOL, "drain_rx_pools %p", dev);
  
  for (pool = 0; pool < NUM_RX_POOLS; ++pool)
    drain_rx_pool (dev, pool);
}

static void fill_rx_pool (amb_dev * dev, unsigned char pool,
                                 gfp_t priority)
{
  rx_in rx;
  amb_rxq * rxq;
  
  PRINTD (DBG_FLOW|DBG_POOL, "fill_rx_pool %p %hu %x", dev, pool, priority);
  
  if (test_bit (dead, &dev->flags))
    return;
  
  rxq = &dev->rxq[pool];
  while (rxq->pending < rxq->maximum && rxq->pending < rxq->buffers_wanted) {
    
    struct sk_buff * skb = alloc_skb (rxq->buffer_size, priority);
    if (!skb) {
      PRINTD (DBG_SKB|DBG_POOL, "failed to allocate skb for RX pool %hu", pool);
      return;
    }
    if (check_area (skb->data, skb->truesize)) {
      dev_kfree_skb_any (skb);
      return;
    }
    // cast needed as there is no %? for pointer differences
    PRINTD (DBG_SKB, "allocated skb at %p, head %p, area %li",
	    skb, skb->head, (long) (skb_end_pointer(skb) - skb->head));
    rx.handle = virt_to_bus (skb);
    rx.host_address = cpu_to_be32 (virt_to_bus (skb->data));
    if (rx_give (dev, &rx, pool))
      dev_kfree_skb_any (skb);
    
  }
  
  return;
}

// top up all RX pools
static void fill_rx_pools (amb_dev * dev) {
  unsigned char pool;
  
  PRINTD (DBG_FLOW|DBG_POOL, "fill_rx_pools %p", dev);
  
  for (pool = 0; pool < NUM_RX_POOLS; ++pool)
    fill_rx_pool (dev, pool, GFP_ATOMIC);
  
  return;
}

/********** enable host interrupts **********/

static void interrupts_on (amb_dev * dev) {
  wr_plain (dev, offsetof(amb_mem, interrupt_control),
	    rd_plain (dev, offsetof(amb_mem, interrupt_control))
	    | AMB_INTERRUPT_BITS);
}

/********** disable host interrupts **********/

static void interrupts_off (amb_dev * dev) {
  wr_plain (dev, offsetof(amb_mem, interrupt_control),
	    rd_plain (dev, offsetof(amb_mem, interrupt_control))
	    &~ AMB_INTERRUPT_BITS);
}

/********** interrupt handling **********/

static irqreturn_t interrupt_handler(int irq, void *dev_id) {
  amb_dev * dev = dev_id;
  
  PRINTD (DBG_IRQ|DBG_FLOW, "interrupt_handler: %p", dev_id);
  
  {
    u32 interrupt = rd_plain (dev, offsetof(amb_mem, interrupt));
  
    // for us or someone else sharing the same interrupt
    if (!interrupt) {
      PRINTD (DBG_IRQ, "irq not for me: %d", irq);
      return IRQ_NONE;
    }
    
    // definitely for us
    PRINTD (DBG_IRQ, "FYI: interrupt was %08x", interrupt);
    wr_plain (dev, offsetof(amb_mem, interrupt), -1);
  }
  
  {
    unsigned int irq_work = 0;
    unsigned char pool;
    for (pool = 0; pool < NUM_RX_POOLS; ++pool)
      while (!rx_take (dev, pool))
	++irq_work;
    while (!tx_take (dev))
      ++irq_work;
  
    if (irq_work) {
      fill_rx_pools (dev);

      PRINTD (DBG_IRQ, "work done: %u", irq_work);
    } else {
      PRINTD (DBG_IRQ|DBG_WARN, "no work done");
    }
  }
  
  PRINTD (DBG_IRQ|DBG_FLOW, "interrupt_handler done: %p", dev_id);
  return IRQ_HANDLED;
}

/********** make rate (not quite as much fun as Horizon) **********/

static int make_rate (unsigned int rate, rounding r,
		      u16 * bits, unsigned int * actual) {
  unsigned char exp = -1; // hush gcc
  unsigned int man = -1;  // hush gcc
  
  PRINTD (DBG_FLOW|DBG_QOS, "make_rate %u", rate);
  
  // rates in cells per second, ITU format (nasty 16-bit floating-point)
  // given 5-bit e and 9-bit m:
  // rate = EITHER (1+m/2^9)*2^e    OR 0
  // bits = EITHER 1<<14 | e<<9 | m OR 0
  // (bit 15 is "reserved", bit 14 "non-zero")
  // smallest rate is 0 (special representation)
  // largest rate is (1+511/512)*2^31 = 4290772992 (< 2^32-1)
  // smallest non-zero rate is (1+0/512)*2^0 = 1 (> 0)
  // simple algorithm:
  // find position of top bit, this gives e
  // remove top bit and shift (rounding if feeling clever) by 9-e
  
  // ucode bug: please don't set bit 14! so 0 rate not representable
  
  if (rate > 0xffc00000U) {
    // larger than largest representable rate
    
    if (r == round_up) {
	return -EINVAL;
    } else {
      exp = 31;
      man = 511;
    }
    
  } else if (rate) {
    // representable rate
    
    exp = 31;
    man = rate;
    
    // invariant: rate = man*2^(exp-31)
    while (!(man & (1<<31))) {
      exp = exp - 1;
      man = man<<1;
    }
    
    // man has top bit set
    // rate = (2^31+(man-2^31))*2^(exp-31)
    // rate = (1+(man-2^31)/2^31)*2^exp
    man = man<<1;
    man &= 0xffffffffU; // a nop on 32-bit systems
    // rate = (1+man/2^32)*2^exp
    
    // exp is in the range 0 to 31, man is in the range 0 to 2^32-1
    // time to lose significance... we want m in the range 0 to 2^9-1
    // rounding presents a minor problem... we first decide which way
    // we are rounding (based on given rounding direction and possibly
    // the bits of the mantissa that are to be discarded).
    
    switch (r) {
      case round_down: {
	// just truncate
	man = man>>(32-9);
	break;
      }
      case round_up: {
	// check all bits that we are discarding
	if (man & (~0U>>9)) {
	  man = (man>>(32-9)) + 1;
	  if (man == (1<<9)) {
	    // no need to check for round up outside of range
	    man = 0;
	    exp += 1;
	  }
	} else {
	  man = (man>>(32-9));
	}
	break;
      }
      case round_nearest: {
	// check msb that we are discarding
	if (man & (1<<(32-9-1))) {
	  man = (man>>(32-9)) + 1;
	  if (man == (1<<9)) {
	    // no need to check for round up outside of range
	    man = 0;
	    exp += 1;
	  }
	} else {
	  man = (man>>(32-9));
	}
	break;
      }
    }
    
  } else {
    // zero rate - not representable
    
    if (r == round_down) {
      return -EINVAL;
    } else {
      exp = 0;
      man = 0;
    }
    
  }
  
  PRINTD (DBG_QOS, "rate: man=%u, exp=%hu", man, exp);
  
  if (bits)
    *bits = /* (1<<14) | */ (exp<<9) | man;
  
  if (actual)
    *actual = (exp >= 9)
      ? (1 << exp) + (man << (exp-9))
      : (1 << exp) + ((man + (1<<(9-exp-1))) >> (9-exp));
  
  return 0;
}

/********** Linux ATM Operations **********/

// some are not yet implemented while others do not make sense for
// this device

/********** Open a VC **********/

static int amb_open (struct atm_vcc * atm_vcc)
{
  int error;
  
  struct atm_qos * qos;
  struct atm_trafprm * txtp;
  struct atm_trafprm * rxtp;
  u16 tx_rate_bits = -1; // hush gcc
  u16 tx_vc_bits = -1; // hush gcc
  u16 tx_frame_bits = -1; // hush gcc
  
  amb_dev * dev = AMB_DEV(atm_vcc->dev);
  amb_vcc * vcc;
  unsigned char pool = -1; // hush gcc
  short vpi = atm_vcc->vpi;
  int vci = atm_vcc->vci;
  
  PRINTD (DBG_FLOW|DBG_VCC, "amb_open %x %x", vpi, vci);
  
#ifdef ATM_VPI_UNSPEC
  // UNSPEC is deprecated, remove this code eventually
  if (vpi == ATM_VPI_UNSPEC || vci == ATM_VCI_UNSPEC) {
    PRINTK (KERN_WARNING, "rejecting open with unspecified VPI/VCI (deprecated)");
    return -EINVAL;
  }
#endif
  
  if (!(0 <= vpi && vpi < (1<<NUM_VPI_BITS) &&
	0 <= vci && vci < (1<<NUM_VCI_BITS))) {
    PRINTD (DBG_WARN|DBG_VCC, "VPI/VCI out of range: %hd/%d", vpi, vci);
    return -EINVAL;
  }
  
  qos = &atm_vcc->qos;
  
  if (qos->aal != ATM_AAL5) {
    PRINTD (DBG_QOS, "AAL not supported");
    return -EINVAL;
  }
  
  // traffic parameters
  
  PRINTD (DBG_QOS, "TX:");
  txtp = &qos->txtp;
  if (txtp->traffic_class != ATM_NONE) {
    switch (txtp->traffic_class) {
      case ATM_UBR: {
	// we take "the PCR" as a rate-cap
	int pcr = atm_pcr_goal (txtp);
	if (!pcr) {
	  // no rate cap
	  tx_rate_bits = 0;
	  tx_vc_bits = TX_UBR;
	  tx_frame_bits = TX_FRAME_NOTCAP;
	} else {
	  rounding r;
	  if (pcr < 0) {
	    r = round_down;
	    pcr = -pcr;
	  } else {
	    r = round_up;
	  }
	  error = make_rate (pcr, r, &tx_rate_bits, NULL);
	  if (error)
	    return error;
	  tx_vc_bits = TX_UBR_CAPPED;
	  tx_frame_bits = TX_FRAME_CAPPED;
	}
	break;
      }
#if 0
      case ATM_ABR: {
	pcr = atm_pcr_goal (txtp);
	PRINTD (DBG_QOS, "pcr goal = %d", pcr);
	break;
      }
#endif
      default: {
	// PRINTD (DBG_QOS, "request for non-UBR/ABR denied");
	PRINTD (DBG_QOS, "request for non-UBR denied");
	return -EINVAL;
      }
    }
    PRINTD (DBG_QOS, "tx_rate_bits=%hx, tx_vc_bits=%hx",
	    tx_rate_bits, tx_vc_bits);
  }
  
  PRINTD (DBG_QOS, "RX:");
  rxtp = &qos->rxtp;
  if (rxtp->traffic_class == ATM_NONE) {
    // do nothing
  } else {
    // choose an RX pool (arranged in increasing size)
    for (pool = 0; pool < NUM_RX_POOLS; ++pool)
      if ((unsigned int) rxtp->max_sdu <= dev->rxq[pool].buffer_size) {
	PRINTD (DBG_VCC|DBG_QOS|DBG_POOL, "chose pool %hu (max_sdu %u <= %u)",
		pool, rxtp->max_sdu, dev->rxq[pool].buffer_size);
	break;
      }
    if (pool == NUM_RX_POOLS) {
      PRINTD (DBG_WARN|DBG_VCC|DBG_QOS|DBG_POOL,
	      "no pool suitable for VC (RX max_sdu %d is too large)",
	      rxtp->max_sdu);
      return -EINVAL;
    }
    
    switch (rxtp->traffic_class) {
      case ATM_UBR: {
	break;
      }
#if 0
      case ATM_ABR: {
	pcr = atm_pcr_goal (rxtp);
	PRINTD (DBG_QOS, "pcr goal = %d", pcr);
	break;
      }
#endif
      default: {
	// PRINTD (DBG_QOS, "request for non-UBR/ABR denied");
	PRINTD (DBG_QOS, "request for non-UBR denied");
	return -EINVAL;
      }
    }
  }
  
  // get space for our vcc stuff
  vcc = kmalloc (sizeof(amb_vcc), GFP_KERNEL);
  if (!vcc) {
    PRINTK (KERN_ERR, "out of memory!");
    return -ENOMEM;
  }
  atm_vcc->dev_data = (void *) vcc;
  
  // no failures beyond this point
  
  // we are not really "immediately before allocating the connection
  // identifier in hardware", but it will just have to do!
  set_bit(ATM_VF_ADDR,&atm_vcc->flags);
  
  if (txtp->traffic_class != ATM_NONE) {
    command cmd;
    
    vcc->tx_frame_bits = tx_frame_bits;
    
    mutex_lock(&dev->vcc_sf);
    if (dev->rxer[vci]) {
      // RXer on the channel already, just modify rate...
      cmd.request = cpu_to_be32 (SRB_MODIFY_VC_RATE);
      cmd.args.modify_rate.vc = cpu_to_be32 (vci);  // vpi 0
      cmd.args.modify_rate.rate = cpu_to_be32 (tx_rate_bits << SRB_RATE_SHIFT);
      while (command_do (dev, &cmd))
	schedule();
      // ... and TX flags, preserving the RX pool
      cmd.request = cpu_to_be32 (SRB_MODIFY_VC_FLAGS);
      cmd.args.modify_flags.vc = cpu_to_be32 (vci);  // vpi 0
      cmd.args.modify_flags.flags = cpu_to_be32
	( (AMB_VCC(dev->rxer[vci])->rx_info.pool << SRB_POOL_SHIFT)
	  | (tx_vc_bits << SRB_FLAGS_SHIFT) );
      while (command_do (dev, &cmd))
	schedule();
    } else {
      // no RXer on the channel, just open (with pool zero)
      cmd.request = cpu_to_be32 (SRB_OPEN_VC);
      cmd.args.open.vc = cpu_to_be32 (vci);  // vpi 0
      cmd.args.open.flags = cpu_to_be32 (tx_vc_bits << SRB_FLAGS_SHIFT);
      cmd.args.open.rate = cpu_to_be32 (tx_rate_bits << SRB_RATE_SHIFT);
      while (command_do (dev, &cmd))
	schedule();
    }
    dev->txer[vci].tx_present = 1;
    mutex_unlock(&dev->vcc_sf);
  }
  
  if (rxtp->traffic_class != ATM_NONE) {
    command cmd;
    
    vcc->rx_info.pool = pool;
    
    mutex_lock(&dev->vcc_sf);
    /* grow RX buffer pool */
    if (!dev->rxq[pool].buffers_wanted)
      dev->rxq[pool].buffers_wanted = rx_lats;
    dev->rxq[pool].buffers_wanted += 1;
    fill_rx_pool (dev, pool, GFP_KERNEL);
    
    if (dev->txer[vci].tx_present) {
      // TXer on the channel already
      // switch (from pool zero) to this pool, preserving the TX bits
      cmd.request = cpu_to_be32 (SRB_MODIFY_VC_FLAGS);
      cmd.args.modify_flags.vc = cpu_to_be32 (vci);  // vpi 0
      cmd.args.modify_flags.flags = cpu_to_be32
	( (pool << SRB_POOL_SHIFT)
	  | (dev->txer[vci].tx_vc_bits << SRB_FLAGS_SHIFT) );
    } else {
      // no TXer on the channel, open the VC (with no rate info)
      cmd.request = cpu_to_be32 (SRB_OPEN_VC);
      cmd.args.open.vc = cpu_to_be32 (vci);  // vpi 0
      cmd.args.open.flags = cpu_to_be32 (pool << SRB_POOL_SHIFT);
      cmd.args.open.rate = cpu_to_be32 (0);
    }
    while (command_do (dev, &cmd))
      schedule();
    // this link allows RX frames through
    dev->rxer[vci] = atm_vcc;
    mutex_unlock(&dev->vcc_sf);
  }
  
  // indicate readiness
  set_bit(ATM_VF_READY,&atm_vcc->flags);
  
  return 0;
}

/********** Close a VC **********/

static void amb_close (struct atm_vcc * atm_vcc) {
  amb_dev * dev = AMB_DEV (atm_vcc->dev);
  amb_vcc * vcc = AMB_VCC (atm_vcc);
  u16 vci = atm_vcc->vci;
  
  PRINTD (DBG_VCC|DBG_FLOW, "amb_close");
  
  // indicate unreadiness
  clear_bit(ATM_VF_READY,&atm_vcc->flags);
  
  // disable TXing
  if (atm_vcc->qos.txtp.traffic_class != ATM_NONE) {
    command cmd;
    
    mutex_lock(&dev->vcc_sf);
    if (dev->rxer[vci]) {
      // RXer still on the channel, just modify rate... XXX not really needed
      cmd.request = cpu_to_be32 (SRB_MODIFY_VC_RATE);
      cmd.args.modify_rate.vc = cpu_to_be32 (vci);  // vpi 0
      cmd.args.modify_rate.rate = cpu_to_be32 (0);
      // ... and clear TX rate flags (XXX to stop RM cell output?), preserving RX pool
    } else {
      // no RXer on the channel, close channel
      cmd.request = cpu_to_be32 (SRB_CLOSE_VC);
      cmd.args.close.vc = cpu_to_be32 (vci); // vpi 0
    }
    dev->txer[vci].tx_present = 0;
    while (command_do (dev, &cmd))
      schedule();
    mutex_unlock(&dev->vcc_sf);
  }
  
  // disable RXing
  if (atm_vcc->qos.rxtp.traffic_class != ATM_NONE) {
    command cmd;
    
    // this is (the?) one reason why we need the amb_vcc struct
    unsigned char pool = vcc->rx_info.pool;
    
    mutex_lock(&dev->vcc_sf);
    if (dev->txer[vci].tx_present) {
      // TXer still on the channel, just go to pool zero XXX not really needed
      cmd.request = cpu_to_be32 (SRB_MODIFY_VC_FLAGS);
      cmd.args.modify_flags.vc = cpu_to_be32 (vci);  // vpi 0
      cmd.args.modify_flags.flags = cpu_to_be32
	(dev->txer[vci].tx_vc_bits << SRB_FLAGS_SHIFT);
    } else {
      // no TXer on the channel, close the VC
      cmd.request = cpu_to_be32 (SRB_CLOSE_VC);
      cmd.args.close.vc = cpu_to_be32 (vci); // vpi 0
    }
    // forget the rxer - no more skbs will be pushed
    if (atm_vcc != dev->rxer[vci])
      PRINTK (KERN_ERR, "%s vcc=%p rxer[vci]=%p",
	      "arghhh! we're going to die!",
	      vcc, dev->rxer[vci]);
    dev->rxer[vci] = NULL;
    while (command_do (dev, &cmd))
      schedule();
    
    /* shrink RX buffer pool */
    dev->rxq[pool].buffers_wanted -= 1;
    if (dev->rxq[pool].buffers_wanted == rx_lats) {
      dev->rxq[pool].buffers_wanted = 0;
      drain_rx_pool (dev, pool);
    }
    mutex_unlock(&dev->vcc_sf);
  }
  
  // free our structure
  kfree (vcc);
  
  // say the VPI/VCI is free again
  clear_bit(ATM_VF_ADDR,&atm_vcc->flags);

  return;
}

/********** Send **********/

static int amb_send (struct atm_vcc * atm_vcc, struct sk_buff * skb) {
  amb_dev * dev = AMB_DEV(atm_vcc->dev);
  amb_vcc * vcc = AMB_VCC(atm_vcc);
  u16 vc = atm_vcc->vci;
  unsigned int tx_len = skb->len;
  unsigned char * tx_data = skb->data;
  tx_simple * tx_descr;
  tx_in tx;
  
  if (test_bit (dead, &dev->flags))
    return -EIO;
  
  PRINTD (DBG_FLOW|DBG_TX, "amb_send vc %x data %p len %u",
	  vc, tx_data, tx_len);
  
  dump_skb (">>>", vc, skb);
  
  if (!dev->txer[vc].tx_present) {
    PRINTK (KERN_ERR, "attempt to send on RX-only VC %x", vc);
    return -EBADFD;
  }
  
  // this is a driver private field so we have to set it ourselves,
  // despite the fact that we are _required_ to use it to check for a
  // pop function
  ATM_SKB(skb)->vcc = atm_vcc;
  
  if (skb->len > (size_t) atm_vcc->qos.txtp.max_sdu) {
    PRINTK (KERN_ERR, "sk_buff length greater than agreed max_sdu, dropping...");
    return -EIO;
  }
  
  if (check_area (skb->data, skb->len)) {
    atomic_inc(&atm_vcc->stats->tx_err);
    return -ENOMEM; // ?
  }
  
  // allocate memory for fragments
  tx_descr = kmalloc (sizeof(tx_simple), GFP_KERNEL);
  if (!tx_descr) {
    PRINTK (KERN_ERR, "could not allocate TX descriptor");
    return -ENOMEM;
  }
  if (check_area (tx_descr, sizeof(tx_simple))) {
    kfree (tx_descr);
    return -ENOMEM;
  }
  PRINTD (DBG_TX, "fragment list allocated at %p", tx_descr);
  
  tx_descr->skb = skb;
  
  tx_descr->tx_frag.bytes = cpu_to_be32 (tx_len);
  tx_descr->tx_frag.address = cpu_to_be32 (virt_to_bus (tx_data));
  
  tx_descr->tx_frag_end.handle = virt_to_bus (tx_descr);
  tx_descr->tx_frag_end.vc = 0;
  tx_descr->tx_frag_end.next_descriptor_length = 0;
  tx_descr->tx_frag_end.next_descriptor = 0;
#ifdef AMB_NEW_MICROCODE
  tx_descr->tx_frag_end.cpcs_uu = 0;
  tx_descr->tx_frag_end.cpi = 0;
  tx_descr->tx_frag_end.pad = 0;
#endif
  
  tx.vc = cpu_to_be16 (vcc->tx_frame_bits | vc);
  tx.tx_descr_length = cpu_to_be16 (sizeof(tx_frag)+sizeof(tx_frag_end));
  tx.tx_descr_addr = cpu_to_be32 (virt_to_bus (&tx_descr->tx_frag));
  
  while (tx_give (dev, &tx))
    schedule();
  return 0;
}

/********** Change QoS on a VC **********/

// int amb_change_qos (struct atm_vcc * atm_vcc, struct atm_qos * qos, int flags);

/********** Free RX Socket Buffer **********/

#if 0
static void amb_free_rx_skb (struct atm_vcc * atm_vcc, struct sk_buff * skb) {
  amb_dev * dev = AMB_DEV (atm_vcc->dev);
  amb_vcc * vcc = AMB_VCC (atm_vcc);
  unsigned char pool = vcc->rx_info.pool;
  rx_in rx;
  
  // This may be unsafe for various reasons that I cannot really guess
  // at. However, I note that the ATM layer calls kfree_skb rather
  // than dev_kfree_skb at this point so we are least covered as far
  // as buffer locking goes. There may be bugs if pcap clones RX skbs.

  PRINTD (DBG_FLOW|DBG_SKB, "amb_rx_free skb %p (atm_vcc %p, vcc %p)",
	  skb, atm_vcc, vcc);
  
  rx.handle = virt_to_bus (skb);
  rx.host_address = cpu_to_be32 (virt_to_bus (skb->data));
  
  skb->data = skb->head;
  skb->tail = skb->head;
  skb->len = 0;
  
  if (!rx_give (dev, &rx, pool)) {
    // success
    PRINTD (DBG_SKB|DBG_POOL, "recycled skb for pool %hu", pool);
    return;
  }
  
  // just do what the ATM layer would have done
  dev_kfree_skb_any (skb);
  
  return;
}
#endif

/********** Proc File Output **********/

static int amb_proc_read (struct atm_dev * atm_dev, loff_t * pos, char * page) {
  amb_dev * dev = AMB_DEV (atm_dev);
  int left = *pos;
  unsigned char pool;
  
  PRINTD (DBG_FLOW, "amb_proc_read");
  
  /* more diagnostics here? */
  
  if (!left--) {
    amb_stats * s = &dev->stats;
    return sprintf (page,
		    "frames: TX OK %lu, RX OK %lu, RX bad %lu "
		    "(CRC %lu, long %lu, aborted %lu, unused %lu).\n",
		    s->tx_ok, s->rx.ok, s->rx.error,
		    s->rx.badcrc, s->rx.toolong,
		    s->rx.aborted, s->rx.unused);
  }
  
  if (!left--) {
    amb_cq * c = &dev->cq;
    return sprintf (page, "cmd queue [cur/hi/max]: %u/%u/%u. ",
		    c->pending, c->high, c->maximum);
  }
  
  if (!left--) {
    amb_txq * t = &dev->txq;
    return sprintf (page, "TX queue [cur/max high full]: %u/%u %u %u.\n",
		    t->pending, t->maximum, t->high, t->filled);
  }
  
  if (!left--) {
    unsigned int count = sprintf (page, "RX queues [cur/max/req low empty]:");
    for (pool = 0; pool < NUM_RX_POOLS; ++pool) {
      amb_rxq * r = &dev->rxq[pool];
      count += sprintf (page+count, " %u/%u/%u %u %u",
			r->pending, r->maximum, r->buffers_wanted, r->low, r->emptied);
    }
    count += sprintf (page+count, ".\n");
    return count;
  }
  
  if (!left--) {
    unsigned int count = sprintf (page, "RX buffer sizes:");
    for (pool = 0; pool < NUM_RX_POOLS; ++pool) {
      amb_rxq * r = &dev->rxq[pool];
      count += sprintf (page+count, " %u", r->buffer_size);
    }
    count += sprintf (page+count, ".\n");
    return count;
  }
  
#if 0
  if (!left--) {
    // suni block etc?
  }
#endif
  
  return 0;
}

/********** Operation Structure **********/

static const struct atmdev_ops amb_ops = {
  .open         = amb_open,
  .close	= amb_close,
  .send         = amb_send,
  .proc_read	= amb_proc_read,
  .owner	= THIS_MODULE,
};

/********** housekeeping **********/
static void do_housekeeping (unsigned long arg) {
  amb_dev * dev = (amb_dev *) arg;
  
  // could collect device-specific (not driver/atm-linux) stats here
      
  // last resort refill once every ten seconds
  fill_rx_pools (dev);
  mod_timer(&dev->housekeeping, jiffies + 10*HZ);
  
  return;
}

/********** creation of communication queues **********/

static int __devinit create_queues (amb_dev * dev, unsigned int cmds,
				 unsigned int txs, unsigned int * rxs,
				 unsigned int * rx_buffer_sizes) {
  unsigned char pool;
  size_t total = 0;
  void * memory;
  void * limit;
  
  PRINTD (DBG_FLOW, "create_queues %p", dev);
  
  total += cmds * sizeof(command);
  
  total += txs * (sizeof(tx_in) + sizeof(tx_out));
  
  for (pool = 0; pool < NUM_RX_POOLS; ++pool)
    total += rxs[pool] * (sizeof(rx_in) + sizeof(rx_out));
  
  memory = kmalloc (total, GFP_KERNEL);
  if (!memory) {
    PRINTK (KERN_ERR, "could not allocate queues");
    return -ENOMEM;
  }
  if (check_area (memory, total)) {
    PRINTK (KERN_ERR, "queues allocated in nasty area");
    kfree (memory);
    return -ENOMEM;
  }
  
  limit = memory + total;
  PRINTD (DBG_INIT, "queues from %p to %p", memory, limit);
  
  PRINTD (DBG_CMD, "command queue at %p", memory);
  
  {
    command * cmd = memory;
    amb_cq * cq = &dev->cq;
    
    cq->pending = 0;
    cq->high = 0;
    cq->maximum = cmds - 1;
    
    cq->ptrs.start = cmd;
    cq->ptrs.in = cmd;
    cq->ptrs.out = cmd;
    cq->ptrs.limit = cmd + cmds;
    
    memory = cq->ptrs.limit;
  }
  
  PRINTD (DBG_TX, "TX queue pair at %p", memory);
  
  {
    tx_in * in = memory;
    tx_out * out;
    amb_txq * txq = &dev->txq;
    
    txq->pending = 0;
    txq->high = 0;
    txq->filled = 0;
    txq->maximum = txs - 1;
    
    txq->in.start = in;
    txq->in.ptr = in;
    txq->in.limit = in + txs;
    
    memory = txq->in.limit;
    out = memory;
    
    txq->out.start = out;
    txq->out.ptr = out;
    txq->out.limit = out + txs;
    
    memory = txq->out.limit;
  }
  
  PRINTD (DBG_RX, "RX queue pairs at %p", memory);
  
  for (pool = 0; pool < NUM_RX_POOLS; ++pool) {
    rx_in * in = memory;
    rx_out * out;
    amb_rxq * rxq = &dev->rxq[pool];
    
    rxq->buffer_size = rx_buffer_sizes[pool];
    rxq->buffers_wanted = 0;
    
    rxq->pending = 0;
    rxq->low = rxs[pool] - 1;
    rxq->emptied = 0;
    rxq->maximum = rxs[pool] - 1;
    
    rxq->in.start = in;
    rxq->in.ptr = in;
    rxq->in.limit = in + rxs[pool];
    
    memory = rxq->in.limit;
    out = memory;
    
    rxq->out.start = out;
    rxq->out.ptr = out;
    rxq->out.limit = out + rxs[pool];
    
    memory = rxq->out.limit;
  }
  
  if (memory == limit) {
    return 0;
  } else {
    PRINTK (KERN_ERR, "bad queue alloc %p != %p (tell maintainer)", memory, limit);
    kfree (limit - total);
    return -ENOMEM;
  }
  
}

/********** destruction of communication queues **********/

static void destroy_queues (amb_dev * dev) {
  // all queues assumed empty
  void * memory = dev->cq.ptrs.start;
  // includes txq.in, txq.out, rxq[].in and rxq[].out
  
  PRINTD (DBG_FLOW, "destroy_queues %p", dev);
  
  PRINTD (DBG_INIT, "freeing queues at %p", memory);
  kfree (memory);
  
  return;
}

/********** basic loader commands and error handling **********/
// centisecond timeouts - guessing away here
static unsigned int command_timeouts [] = {
	[host_memory_test]     = 15,
	[read_adapter_memory]  = 2,
	[write_adapter_memory] = 2,
	[adapter_start]        = 50,
	[get_version_number]   = 10,
	[interrupt_host]       = 1,
	[flash_erase_sector]   = 1,
	[adap_download_block]  = 1,
	[adap_erase_flash]     = 1,
	[adap_run_in_iram]     = 1,
	[adap_end_download]    = 1
};


static unsigned int command_successes [] = {
	[host_memory_test]     = COMMAND_PASSED_TEST,
	[read_adapter_memory]  = COMMAND_READ_DATA_OK,
	[write_adapter_memory] = COMMAND_WRITE_DATA_OK,
	[adapter_start]        = COMMAND_COMPLETE,
	[get_version_number]   = COMMAND_COMPLETE,
	[interrupt_host]       = COMMAND_COMPLETE,
	[flash_erase_sector]   = COMMAND_COMPLETE,
	[adap_download_block]  = COMMAND_COMPLETE,
	[adap_erase_flash]     = COMMAND_COMPLETE,
	[adap_run_in_iram]     = COMMAND_COMPLETE,
	[adap_end_download]    = COMMAND_COMPLETE
};
  
static  int decode_loader_result (loader_command cmd, u32 result)
{
	int res;
	const char *msg;

	if (result == command_successes[cmd])
		return 0;

	switch (result) {
		case BAD_COMMAND:
			res = -EINVAL;
			msg = "bad command";
			break;
		case COMMAND_IN_PROGRESS:
			res = -ETIMEDOUT;
			msg = "command in progress";
			break;
		case COMMAND_PASSED_TEST:
			res = 0;
			msg = "command passed test";
			break;
		case COMMAND_FAILED_TEST:
			res = -EIO;
			msg = "command failed test";
			break;
		case COMMAND_READ_DATA_OK:
			res = 0;
			msg = "command read data ok";
			break;
		case COMMAND_READ_BAD_ADDRESS:
			res = -EINVAL;
			msg = "command read bad address";
			break;
		case COMMAND_WRITE_DATA_OK:
			res = 0;
			msg = "command write data ok";
			break;
		case COMMAND_WRITE_BAD_ADDRESS:
			res = -EINVAL;
			msg = "command write bad address";
			break;
		case COMMAND_WRITE_FLASH_FAILURE:
			res = -EIO;
			msg = "command write flash failure";
			break;
		case COMMAND_COMPLETE:
			res = 0;
			msg = "command complete";
			break;
		case COMMAND_FLASH_ERASE_FAILURE:
			res = -EIO;
			msg = "command flash erase failure";
			break;
		case COMMAND_WRITE_BAD_DATA:
			res = -EINVAL;
			msg = "command write bad data";
			break;
		default:
			res = -EINVAL;
			msg = "unknown error";
			PRINTD (DBG_LOAD|DBG_ERR,
				"decode_loader_result got %d=%x !",
				result, result);
			break;
	}

	PRINTK (KERN_ERR, "%s", msg);
	return res;
}

static int __devinit do_loader_command (volatile loader_block * lb,
				     const amb_dev * dev, loader_command cmd) {
  
  unsigned long timeout;
  
  PRINTD (DBG_FLOW|DBG_LOAD, "do_loader_command");
  
  /* do a command
     
     Set the return value to zero, set the command type and set the
     valid entry to the right magic value. The payload is already
     correctly byte-ordered so we leave it alone. Hit the doorbell
     with the bus address of this structure.
     
  */
  
  lb->result = 0;
  lb->command = cpu_to_be32 (cmd);
  lb->valid = cpu_to_be32 (DMA_VALID);
  // dump_registers (dev);
  // dump_loader_block (lb);
  wr_mem (dev, offsetof(amb_mem, doorbell), virt_to_bus (lb) & ~onegigmask);
  
  timeout = command_timeouts[cmd] * 10;
  
  while (!lb->result || lb->result == cpu_to_be32 (COMMAND_IN_PROGRESS))
    if (timeout) {
      timeout = msleep_interruptible(timeout);
    } else {
      PRINTD (DBG_LOAD|DBG_ERR, "command %d timed out", cmd);
      dump_registers (dev);
      dump_loader_block (lb);
      return -ETIMEDOUT;
    }
  
  if (cmd == adapter_start) {
    // wait for start command to acknowledge...
    timeout = 100;
    while (rd_plain (dev, offsetof(amb_mem, doorbell)))
      if (timeout) {
	timeout = msleep_interruptible(timeout);
      } else {
	PRINTD (DBG_LOAD|DBG_ERR, "start command did not clear doorbell, res=%08x",
		be32_to_cpu (lb->result));
	dump_registers (dev);
	return -ETIMEDOUT;
      }
    return 0;
  } else {
    return decode_loader_result (cmd, be32_to_cpu (lb->result));
  }
  
}

/* loader: determine loader version */

static int __devinit get_loader_version (loader_block * lb,
				      const amb_dev * dev, u32 * version) {
  int res;
  
  PRINTD (DBG_FLOW|DBG_LOAD, "get_loader_version");
  
  res = do_loader_command (lb, dev, get_version_number);
  if (res)
    return res;
  if (version)
    *version = be32_to_cpu (lb->payload.version);
  return 0;
}

/* loader: write memory data blocks */

static int __devinit loader_write (loader_block* lb,
				   const amb_dev *dev,
				   const struct ihex_binrec *rec) {
  transfer_block * tb = &lb->payload.transfer;
  
  PRINTD (DBG_FLOW|DBG_LOAD, "loader_write");

  tb->address = rec->addr;
  tb->count = cpu_to_be32(be16_to_cpu(rec->len) / 4);
  memcpy(tb->data, rec->data, be16_to_cpu(rec->len));
  return do_loader_command (lb, dev, write_adapter_memory);
}

/* loader: verify memory data blocks */

static int __devinit loader_verify (loader_block * lb,
				    const amb_dev *dev,
				    const struct ihex_binrec *rec) {
  transfer_block * tb = &lb->payload.transfer;
  int res;
  
  PRINTD (DBG_FLOW|DBG_LOAD, "loader_verify");
  
  tb->address = rec->addr;
  tb->count = cpu_to_be32(be16_to_cpu(rec->len) / 4);
  res = do_loader_command (lb, dev, read_adapter_memory);
  if (!res && memcmp(tb->data, rec->data, be16_to_cpu(rec->len)))
    res = -EINVAL;
  return res;
}

/* loader: start microcode */

static int __devinit loader_start (loader_block * lb,
				const amb_dev * dev, u32 address) {
  PRINTD (DBG_FLOW|DBG_LOAD, "loader_start");
  
  lb->payload.start = cpu_to_be32 (address);
  return do_loader_command (lb, dev, adapter_start);
}

/********** reset card **********/

static inline void sf (const char * msg)
{
	PRINTK (KERN_ERR, "self-test failed: %s", msg);
}

static int amb_reset (amb_dev * dev, int diags) {
  u32 word;
  
  PRINTD (DBG_FLOW|DBG_LOAD, "amb_reset");
  
  word = rd_plain (dev, offsetof(amb_mem, reset_control));
  // put card into reset state
  wr_plain (dev, offsetof(amb_mem, reset_control), word | AMB_RESET_BITS);
  // wait a short while
  udelay (10);
#if 1
  // put card into known good state
  wr_plain (dev, offsetof(amb_mem, interrupt_control), AMB_DOORBELL_BITS);
  // clear all interrupts just in case
  wr_plain (dev, offsetof(amb_mem, interrupt), -1);
#endif
  // clear self-test done flag
  wr_plain (dev, offsetof(amb_mem, mb.loader.ready), 0);
  // take card out of reset state
  wr_plain (dev, offsetof(amb_mem, reset_control), word &~ AMB_RESET_BITS);
  
  if (diags) { 
    unsigned long timeout;
    // 4.2 second wait
    msleep(4200);
    // half second time-out
    timeout = 500;
    while (!rd_plain (dev, offsetof(amb_mem, mb.loader.ready)))
      if (timeout) {
	timeout = msleep_interruptible(timeout);
      } else {
	PRINTD (DBG_LOAD|DBG_ERR, "reset timed out");
	return -ETIMEDOUT;
      }
    
    // get results of self-test
    // XXX double check byte-order
    word = rd_mem (dev, offsetof(amb_mem, mb.loader.result));
    if (word & SELF_TEST_FAILURE) {
      if (word & GPINT_TST_FAILURE)
	sf ("interrupt");
      if (word & SUNI_DATA_PATTERN_FAILURE)
	sf ("SUNI data pattern");
      if (word & SUNI_DATA_BITS_FAILURE)
	sf ("SUNI data bits");
      if (word & SUNI_UTOPIA_FAILURE)
	sf ("SUNI UTOPIA interface");
      if (word & SUNI_FIFO_FAILURE)
	sf ("SUNI cell buffer FIFO");
      if (word & SRAM_FAILURE)
	sf ("bad SRAM");
      // better return value?
      return -EIO;
    }
    
  }
  return 0;
}

/********** transfer and start the microcode **********/

static int __devinit ucode_init (loader_block * lb, amb_dev * dev) {
  const struct firmware *fw;
  unsigned long start_address;
  const struct ihex_binrec *rec;
  const char *errmsg = 0;
  int res;

  res = request_ihex_firmware(&fw, "atmsar11.fw", &dev->pci_dev->dev);
  if (res) {
    PRINTK (KERN_ERR, "Cannot load microcode data");
    return res;
  }

  /* First record contains just the start address */
  rec = (const struct ihex_binrec *)fw->data;
  if (be16_to_cpu(rec->len) != sizeof(__be32) || be32_to_cpu(rec->addr)) {
    errmsg = "no start record";
    goto fail;
  }
  start_address = be32_to_cpup((__be32 *)rec->data);

  rec = ihex_next_binrec(rec);

  PRINTD (DBG_FLOW|DBG_LOAD, "ucode_init");

  while (rec) {
    PRINTD (DBG_LOAD, "starting region (%x, %u)", be32_to_cpu(rec->addr),
	    be16_to_cpu(rec->len));
    if (be16_to_cpu(rec->len) > 4 * MAX_TRANSFER_DATA) {
	    errmsg = "record too long";
	    goto fail;
    }
    if (be16_to_cpu(rec->len) & 3) {
	    errmsg = "odd number of bytes";
	    goto fail;
    }
    res = loader_write(lb, dev, rec);
    if (res)
      break;

    res = loader_verify(lb, dev, rec);
    if (res)
      break;
  }
  release_firmware(fw);
  if (!res)
    res = loader_start(lb, dev, start_address);

  return res;
fail:
  release_firmware(fw);
  PRINTK(KERN_ERR, "Bad microcode data (%s)", errmsg);
  return -EINVAL;
}

/********** give adapter parameters **********/
  
static inline __be32 bus_addr(void * addr) {
    return cpu_to_be32 (virt_to_bus (addr));
}

static int __devinit amb_talk (amb_dev * dev) {
  adap_talk_block a;
  unsigned char pool;
  unsigned long timeout;
  
  PRINTD (DBG_FLOW, "amb_talk %p", dev);
  
  a.command_start = bus_addr (dev->cq.ptrs.start);
  a.command_end   = bus_addr (dev->cq.ptrs.limit);
  a.tx_start      = bus_addr (dev->txq.in.start);
  a.tx_end        = bus_addr (dev->txq.in.limit);
  a.txcom_start   = bus_addr (dev->txq.out.start);
  a.txcom_end     = bus_addr (dev->txq.out.limit);
  
  for (pool = 0; pool < NUM_RX_POOLS; ++pool) {
    // the other "a" items are set up by the adapter
    a.rec_struct[pool].buffer_start = bus_addr (dev->rxq[pool].in.start);
    a.rec_struct[pool].buffer_end   = bus_addr (dev->rxq[pool].in.limit);
    a.rec_struct[pool].rx_start     = bus_addr (dev->rxq[pool].out.start);
    a.rec_struct[pool].rx_end       = bus_addr (dev->rxq[pool].out.limit);
    a.rec_struct[pool].buffer_size = cpu_to_be32 (dev->rxq[pool].buffer_size);
  }
  
#ifdef AMB_NEW_MICROCODE
  // disable fast PLX prefetching
  a.init_flags = 0;
#endif
  
  // pass the structure
  wr_mem (dev, offsetof(amb_mem, doorbell), virt_to_bus (&a));
  
  // 2.2 second wait (must not touch doorbell during 2 second DMA test)
  msleep(2200);
  // give the adapter another half second?
  timeout = 500;
  while (rd_plain (dev, offsetof(amb_mem, doorbell)))
    if (timeout) {
      timeout = msleep_interruptible(timeout);
    } else {
      PRINTD (DBG_INIT|DBG_ERR, "adapter init timed out");
      return -ETIMEDOUT;
    }
  
  return 0;
}

// get microcode version
static void __devinit amb_ucode_version (amb_dev * dev) {
  u32 major;
  u32 minor;
  command cmd;
  cmd.request = cpu_to_be32 (SRB_GET_VERSION);
  while (command_do (dev, &cmd)) {
    set_current_state(TASK_UNINTERRUPTIBLE);
    schedule();
  }
  major = be32_to_cpu (cmd.args.version.major);
  minor = be32_to_cpu (cmd.args.version.minor);
  PRINTK (KERN_INFO, "microcode version is %u.%u", major, minor);
}
  
// get end station address
static void __devinit amb_esi (amb_dev * dev, u8 * esi) {
  u32 lower4;
  u16 upper2;
  command cmd;
  
  cmd.request = cpu_to_be32 (SRB_GET_BIA);
  while (command_do (dev, &cmd)) {
    set_current_state(TASK_UNINTERRUPTIBLE);
    schedule();
  }
  lower4 = be32_to_cpu (cmd.args.bia.lower4);
  upper2 = be32_to_cpu (cmd.args.bia.upper2);
  PRINTD (DBG_LOAD, "BIA: lower4: %08x, upper2 %04x", lower4, upper2);
  
  if (esi) {
    unsigned int i;
    
    PRINTDB (DBG_INIT, "ESI:");
    for (i = 0; i < ESI_LEN; ++i) {
      if (i < 4)
	  esi[i] = bitrev8(lower4>>(8*i));
      else
	  esi[i] = bitrev8(upper2>>(8*(i-4)));
      PRINTDM (DBG_INIT, " %02x", esi[i]);
    }
    
    PRINTDE (DBG_INIT, "");
  }
  
  return;
}
  
static void fixup_plx_window (amb_dev *dev, loader_block *lb)
{
	// fix up the PLX-mapped window base address to match the block
	unsigned long blb;
	u32 mapreg;
	blb = virt_to_bus(lb);
	// the kernel stack had better not ever cross a 1Gb boundary!
	mapreg = rd_plain (dev, offsetof(amb_mem, stuff[10]));
	mapreg &= ~onegigmask;
	mapreg |= blb & onegigmask;
	wr_plain (dev, offsetof(amb_mem, stuff[10]), mapreg);
	return;
}

static int __devinit amb_init (amb_dev * dev)
{
  loader_block lb;
  
  u32 version;
  
  if (amb_reset (dev, 1)) {
    PRINTK (KERN_ERR, "card reset failed!");
  } else {
    fixup_plx_window (dev, &lb);
    
    if (get_loader_version (&lb, dev, &version)) {
      PRINTK (KERN_INFO, "failed to get loader version");
    } else {
      PRINTK (KERN_INFO, "loader version is %08x", version);
      
      if (ucode_init (&lb, dev)) {
	PRINTK (KERN_ERR, "microcode failure");
      } else if (create_queues (dev, cmds, txs, rxs, rxs_bs)) {
	PRINTK (KERN_ERR, "failed to get memory for queues");
      } else {
	
	if (amb_talk (dev)) {
	  PRINTK (KERN_ERR, "adapter did not accept queues");
	} else {
	  
	  amb_ucode_version (dev);
	  return 0;
	  
	} /* amb_talk */
	
	destroy_queues (dev);
      } /* create_queues, ucode_init */
      
      amb_reset (dev, 0);
    } /* get_loader_version */
    
  } /* amb_reset */
  
  return -EINVAL;
}

static void setup_dev(amb_dev *dev, struct pci_dev *pci_dev) 
{
      unsigned char pool;
      
      // set up known dev items straight away
      dev->pci_dev = pci_dev; 
      pci_set_drvdata(pci_dev, dev);
      
      dev->iobase = pci_resource_start (pci_dev, 1);
      dev->irq = pci_dev->irq; 
      dev->membase = bus_to_virt(pci_resource_start(pci_dev, 0));
      
      // flags (currently only dead)
      dev->flags = 0;
      
      // Allocate cell rates (fibre)
      // ATM_OC3_PCR = 1555200000/8/270*260/53 - 29/53
      // to be really pedantic, this should be ATM_OC3c_PCR
      dev->tx_avail = ATM_OC3_PCR;
      dev->rx_avail = ATM_OC3_PCR;
      
      // semaphore for txer/rxer modifications - we cannot use a
      // spinlock as the critical region needs to switch processes
      mutex_init(&dev->vcc_sf);
      // queue manipulation spinlocks; we want atomic reads and
      // writes to the queue descriptors (handles IRQ and SMP)
      // consider replacing "int pending" -> "atomic_t available"
      // => problem related to who gets to move queue pointers
      spin_lock_init (&dev->cq.lock);
      spin_lock_init (&dev->txq.lock);
      for (pool = 0; pool < NUM_RX_POOLS; ++pool)
	spin_lock_init (&dev->rxq[pool].lock);
}

static void setup_pci_dev(struct pci_dev *pci_dev)
{
	unsigned char lat;
      
	// enable bus master accesses
	pci_set_master(pci_dev);

	// frobnicate latency (upwards, usually)
	pci_read_config_byte (pci_dev, PCI_LATENCY_TIMER, &lat);

	if (!pci_lat)
		pci_lat = (lat < MIN_PCI_LATENCY) ? MIN_PCI_LATENCY : lat;

	if (lat != pci_lat) {
		PRINTK (KERN_INFO, "Changing PCI latency timer from %hu to %hu",
			lat, pci_lat);
		pci_write_config_byte(pci_dev, PCI_LATENCY_TIMER, pci_lat);
	}
}

static int __devinit amb_probe(struct pci_dev *pci_dev, const struct pci_device_id *pci_ent)
{
	amb_dev * dev;
	int err;
	unsigned int irq;
      
	err = pci_enable_device(pci_dev);
	if (err < 0) {
		PRINTK (KERN_ERR, "skipped broken (PLX rev 2) card");
		goto out;
	}

	// read resources from PCI configuration space
	irq = pci_dev->irq;

	if (pci_dev->device == PCI_DEVICE_ID_MADGE_AMBASSADOR_BAD) {
		PRINTK (KERN_ERR, "skipped broken (PLX rev 2) card");
		err = -EINVAL;
		goto out_disable;
	}

	PRINTD (DBG_INFO, "found Madge ATM adapter (amb) at"
		" IO %llx, IRQ %u, MEM %p",
		(unsigned long long)pci_resource_start(pci_dev, 1),
		irq, bus_to_virt(pci_resource_start(pci_dev, 0)));

	// check IO region
	err = pci_request_region(pci_dev, 1, DEV_LABEL);
	if (err < 0) {
		PRINTK (KERN_ERR, "IO range already in use!");
		goto out_disable;
	}

	dev = kzalloc(sizeof(amb_dev), GFP_KERNEL);
	if (!dev) {
		PRINTK (KERN_ERR, "out of memory!");
		err = -ENOMEM;
		goto out_release;
	}

	setup_dev(dev, pci_dev);

	err = amb_init(dev);
	if (err < 0) {
		PRINTK (KERN_ERR, "adapter initialisation failure");
		goto out_free;
	}

	setup_pci_dev(pci_dev);

	// grab (but share) IRQ and install handler
	err = request_irq(irq, interrupt_handler, IRQF_SHARED, DEV_LABEL, dev);
	if (err < 0) {
		PRINTK (KERN_ERR, "request IRQ failed!");
		goto out_reset;
	}

	dev->atm_dev = atm_dev_register (DEV_LABEL, &pci_dev->dev, &amb_ops, -1,
					 NULL);
	if (!dev->atm_dev) {
		PRINTD (DBG_ERR, "failed to register Madge ATM adapter");
		err = -EINVAL;
		goto out_free_irq;
	}

	PRINTD (DBG_INFO, "registered Madge ATM adapter (no. %d) (%p) at %p",
		dev->atm_dev->number, dev, dev->atm_dev);
		dev->atm_dev->dev_data = (void *) dev;

	// register our address
	amb_esi (dev, dev->atm_dev->esi);

	// 0 bits for vpi, 10 bits for vci
	dev->atm_dev->ci_range.vpi_bits = NUM_VPI_BITS;
	dev->atm_dev->ci_range.vci_bits = NUM_VCI_BITS;

	init_timer(&dev->housekeeping);
	dev->housekeeping.function = do_housekeeping;
	dev->housekeeping.data = (unsigned long) dev;
	mod_timer(&dev->housekeeping, jiffies);

	// enable host interrupts
	interrupts_on (dev);

out:
	return err;

out_free_irq:
	free_irq(irq, dev);
out_reset:
	amb_reset(dev, 0);
out_free:
	kfree(dev);
out_release:
	pci_release_region(pci_dev, 1);
out_disable:
	pci_disable_device(pci_dev);
	goto out;
}


static void __devexit amb_remove_one(struct pci_dev *pci_dev)
{
	struct amb_dev *dev;

	dev = pci_get_drvdata(pci_dev);

	PRINTD(DBG_INFO|DBG_INIT, "closing %p (atm_dev = %p)", dev, dev->atm_dev);
	del_timer_sync(&dev->housekeeping);
	// the drain should not be necessary
	drain_rx_pools(dev);
	interrupts_off(dev);
	amb_reset(dev, 0);
	free_irq(dev->irq, dev);
	pci_disable_device(pci_dev);
	destroy_queues(dev);
	atm_dev_deregister(dev->atm_dev);
	kfree(dev);
	pci_release_region(pci_dev, 1);
}

static void __init amb_check_args (void) {
  unsigned char pool;
  unsigned int max_rx_size;
  
#ifdef DEBUG_AMBASSADOR
  PRINTK (KERN_NOTICE, "debug bitmap is %hx", debug &= DBG_MASK);
#else
  if (debug)
    PRINTK (KERN_NOTICE, "no debugging support");
#endif
  
  if (cmds < MIN_QUEUE_SIZE)
    PRINTK (KERN_NOTICE, "cmds has been raised to %u",
	    cmds = MIN_QUEUE_SIZE);
  
  if (txs < MIN_QUEUE_SIZE)
    PRINTK (KERN_NOTICE, "txs has been raised to %u",
	    txs = MIN_QUEUE_SIZE);
  
  for (pool = 0; pool < NUM_RX_POOLS; ++pool)
    if (rxs[pool] < MIN_QUEUE_SIZE)
      PRINTK (KERN_NOTICE, "rxs[%hu] has been raised to %u",
	      pool, rxs[pool] = MIN_QUEUE_SIZE);
  
  // buffers sizes should be greater than zero and strictly increasing
  max_rx_size = 0;
  for (pool = 0; pool < NUM_RX_POOLS; ++pool)
    if (rxs_bs[pool] <= max_rx_size)
      PRINTK (KERN_NOTICE, "useless pool (rxs_bs[%hu] = %u)",
	      pool, rxs_bs[pool]);
    else
      max_rx_size = rxs_bs[pool];
  
  if (rx_lats < MIN_RX_BUFFERS)
    PRINTK (KERN_NOTICE, "rx_lats has been raised to %u",
	    rx_lats = MIN_RX_BUFFERS);
  
  return;
}

/********** module stuff **********/

MODULE_AUTHOR(maintainer_string);
MODULE_DESCRIPTION(description_string);
MODULE_LICENSE("GPL");
MODULE_FIRMWARE("atmsar11.fw");
module_param(debug,   ushort, 0644);
module_param(cmds,    uint, 0);
module_param(txs,     uint, 0);
module_param_array(rxs,     uint, NULL, 0);
module_param_array(rxs_bs,  uint, NULL, 0);
module_param(rx_lats, uint, 0);
module_param(pci_lat, byte, 0);
MODULE_PARM_DESC(debug,   "debug bitmap, see .h file");
MODULE_PARM_DESC(cmds,    "number of command queue entries");
MODULE_PARM_DESC(txs,     "number of TX queue entries");
MODULE_PARM_DESC(rxs,     "number of RX queue entries [" __MODULE_STRING(NUM_RX_POOLS) "]");
MODULE_PARM_DESC(rxs_bs,  "size of RX buffers [" __MODULE_STRING(NUM_RX_POOLS) "]");
MODULE_PARM_DESC(rx_lats, "number of extra buffers to cope with RX latencies");
MODULE_PARM_DESC(pci_lat, "PCI latency in bus cycles");

/********** module entry **********/

static struct pci_device_id amb_pci_tbl[] = {
	{ PCI_VDEVICE(MADGE, PCI_DEVICE_ID_MADGE_AMBASSADOR), 0 },
	{ PCI_VDEVICE(MADGE, PCI_DEVICE_ID_MADGE_AMBASSADOR_BAD), 0 },
	{ 0, }
};

MODULE_DEVICE_TABLE(pci, amb_pci_tbl);

static struct pci_driver amb_driver = {
	.name =		"amb",
	.probe =	amb_probe,
	.remove =	__devexit_p(amb_remove_one),
	.id_table =	amb_pci_tbl,
};

static int __init amb_module_init (void)
{
  PRINTD (DBG_FLOW|DBG_INIT, "init_module");
  
  // sanity check - cast needed as printk does not support %Zu
  if (sizeof(amb_mem) != 4*16 + 4*12) {
    PRINTK (KERN_ERR, "Fix amb_mem (is %lu words).",
	    (unsigned long) sizeof(amb_mem));
    return -ENOMEM;
  }
  
  show_version();
  
  amb_check_args();
  
  // get the juice
  return pci_register_driver(&amb_driver);
}

/********** module exit **********/

static void __exit amb_module_exit (void)
{
  PRINTD (DBG_FLOW|DBG_INIT, "cleanup_module");

  pci_unregister_driver(&amb_driver);
}

module_init(amb_module_init);
module_exit(amb_module_exit);